CUDA terminal-fence candidate: complete fixed-matrix decision

The complete candidate is inconclusive under the unchanged 5% policy. Retain the production CUDA baseline and the expanded completion test coverage; do not adopt a selected size, backend or mode, retry the trial, or correct by framework controls.

Measured baseline source: 84c88c98c895103395e1b6d6e98f7618848a14d3 (tree 57a755d537e516b91c70c1b555f2803753389862). Measured candidate source: 32fd10a941e783c703d6cc971bf7f18b0cc71c7a (tree e97ff69a80c3ab8ae1d74fc9da8e76940ee7fea2).

The candidate removes one final stream fence after the source-matched final scalar/status readback. Resource ownership, exception completion guard, DeviceGuard, arithmetic, workspace sizes and explicit backend/mode selection remain unchanged. Source identities describe the measured candidates even if production later retains the baseline.

Protocol and interpretation

All 32 original CUDA configurations and all 336 primary case/implementation rows are retained, including 128 PyTorch CUDA controls, 128 cuda-native rows and 80 cuda-cublas rows. Two windows execute ABBA and BAAB in fresh worker processes; window two reverses configuration order. There are 512 raw artifacts and 5,376 stored candidate-case outcomes. Every side has eight process blocks, each with three warmups, 36 Williams-balanced rounds and one repetition. Each row therefore contributes 288 measured rounds per side. The boundary pause is 60 seconds; telemetry is captured only before/after each child.

The primary estimate is the median of all normalized rounds pooled across the eight blocks on each side, then current/baseline. It is not the median of eight process medians. The original comparator resamples artifact blocks and then rounds (10,000 resamples, seed 20261018, 95% percentile interval). Upper bound <= 1.05 supports no material regression; lower bound > 1.05 is a regression; all overlap is inconclusive. Every primary row, including controls, participates in the overall decision. No filtering, retry, pooling of earlier trials, control correction, work normalization or partial mode/size adoption is applied.

Fresh API constructs resources per call. Session reuses explicitly owned CUDA resources. Prepared reuses prepared solve state. Their measurements and verdicts are reported separately; scope medians are never subtracted to infer component costs.

All original primary rows

Rows

Supported <=5%

Inconclusive

Regression

Overall

336

117

219

0

inconclusive

Scope, backend and retained control verdicts

Scope

Implementation

Role

Rows

Supported

Inconclusive

Regression

api

cuda-cublas

native

32

5

27

0

api

cuda-native

native

32

8

24

0

api

torch-cuda-eager

control

32

6

26

0

prepared

cuda-cublas

native

32

15

17

0

prepared

cuda-native

native

80

46

34

0

prepared

torch-cuda-eager

control

80

37

43

0

session

cuda-cublas

native

16

0

16

0

session

cuda-native

native

16

0

16

0

session

torch-cuda-eager

control

16

0

16

0

Numerical and completion evidence

A separate, complete 64-process CUDA preflight finished before timing: 672 candidate-case rows, two independent residual checks per row (1,344 checks), and exact same-candidate iterations/matvecs/restarts and thresholds across both sources and repeated reuse. Timing artifacts retain candidate/case correctness summaries from the collector’s checks outside timing before/after collection. Every retained summary passes its independent true-residual gate and matches the preflight fixture/work identity; individual timed sample invocations are not separately residual-instrumented. Convergence and work are audited separately; no iteration normalization is used.

Both actual C++ CUDA test binaries passed memcheck with full leak checking and synccheck: each run executed 37 tests, with 36 passing and the one expected multi-GPU skip on this host. The candidate completion test queries a borrowed nondefault nonblocking stream immediately after solve, before helper synchronization, download, reset or destruction. Its 184 solves span terminal iteration limits 0/1, initially converged, negative curvature, nonfinite inputs, restart paths, host/device execution, native/cuBLAS, fusion and supported algorithm/preconditioner combinations. Exception synchronization and borrowed resource lifetime remain covered.

Sixteen fixed same-call captures pair ten observed calls after three warmups for cuBLAS Jacobi n=1024 fresh API/Session/prepared in host/device mode, plus native Jacobi fresh API host/device. All 80 paired calls show exactly one fewer cudaStreamSynchronize, cudaGetDevice and cudaSetDevice; launch/memcpy/memset API counts and physical GPU kernel/copy inventories match. The final source-matched copy has two subsequent synchronizations in A and one in B. The analysis retains exact API table/name/thread records, inclusive/deepest phase ownership, unassigned events and physical event context. The D2H designation is source-matched ordering inference rather than decoded API arguments; timestamp containment attributes physical GPU events. Instrumented timings are not framework scores.

Safety checks retained for both sources

Check

Source

Tests

Passed

Expected skips

baseline-memcheck

baseline

37

36

1

baseline-synccheck

baseline

37

36

1

current-memcheck

current

37

36

1

current-synccheck

current

37

36

1

CPU candidate disposition

The independent CPU eight-accumulator candidate abd655a was rejected before timing. Dense-64 native changed stable I/M/R from 49/51/0 to 48/50/0 in both untimed repeats; all independent true-residual gates passed. This is a fixed-work contract rejection, not measured evidence of a slowdown. All three CPU timing plans remain pristine and cancelled. No CPU size/mode fallback, timing retry or work correction was used; production retains the CPU baseline.

Descriptive comparison with PyTorch CUDA

The following counts use native/PyTorch pooled medians from the same artifacts. A ratio above one means a higher native median. They are descriptive, with no new interval or adoption score, and do not adjust the primary A/B comparison. Complete 208 native/framework rows, all eight matched process-block ratios and both implementations’ I/M/R are retained in JSON; cross-framework work differences are explicitly identified.

Native rows with higher medians than the PyTorch control

Scope

Backend

Rows

A pooled slower

B pooled slower

A all 8 blocks slower

B all 8 blocks slower

api

cuda-native

32

0

0

0

0

api

cuda-cublas

32

16

16

16

16

session

cuda-native

16

0

0

0

0

session

cuda-cublas

16

0

0

0

0

prepared

cuda-native

80

0

0

0

0

prepared

cuda-cublas

32

0

0

0

0

Complete primary results

All medians below are nanoseconds per call. A/B block lists retain the original chronological eight process medians on each side; they do not replace the pooled estimator. I/M/R is the exact shared same-candidate numerical work. Stored numeric precision and the comparator’s original per-row intervals are retained without rebootstrap. No rows are omitted.

cuda-release-cuda-dense-f64-v1-none-api-fused1-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

70/73/1

12568480.0

12404465.5

0.9869503313049788

0.8837694480983705

1.0846217393764435

inconclusive

11408434.0 | 11813956.0 | 11465425.5 | 12218891.0 | 14255306.0 | 14576196.5 | 14080129.5 | 13278713.5

11846777.0 | 11323817.0 | 11613471.0 | 12194773.0 | 13273913.0 | 12905990.5 | 13751596.0 | 13196820.0

dense-1024

cuda-native

70/73/1

12144958.0

12151988.0

1.0005788410301624

0.8887206005521764

1.0941604120403488

inconclusive

11344886.5 | 11554106.0 | 11236870.0 | 11750920.0 | 13786573.5 | 13716647.5 | 13661511.0 | 12842823.0

11626146.0 | 11212030.0 | 11396380.5 | 11645276.0 | 12970153.0 | 13196773.5 | 13493697.5 | 12706212.0

dense-1024

torch-cuda-eager

70/73/1

18183710.5

18032091.5

0.9916618228166358

0.8720888163602749

1.0985196563101145

inconclusive

16621032.5 | 16792068.5 | 15979384.5 | 17193781.0 | 21262131.5 | 21341515.5 | 20320887.0 | 18780524.0

17416136.5 | 15796422.0 | 16054689.0 | 17331578.0 | 18952443.5 | 18990563.5 | 19376333.0 | 19626077.0

dense-2048

cuda-cublas

70/73/1

18886969.0

18721901.5

0.9912602440338627

0.9381508278075907

1.0535046829278698

inconclusive

17949890.0 | 18030126.5 | 18018219.0 | 18921595.5 | 20007534.0 | 20478948.0 | 19036776.5 | 19733184.5

17923072.5 | 18289762.0 | 18398964.0 | 17984040.0 | 19259716.5 | 19741626.5 | 19812842.5 | 19628276.0

dense-2048

cuda-native

70/73/1

18451653.0

18207057.5

0.9867439789811785

0.9234752704835729

1.0459341684988264

supported_no_material_regression

17793651.5 | 17596572.0 | 17606557.5 | 18489554.0 | 19577924.0 | 20207994.5 | 19105019.5 | 19388629.0

17563129.5 | 17801830.0 | 17789789.0 | 17570785.0 | 19077272.5 | 19129489.5 | 19359534.5 | 19166742.5

dense-2048

torch-cuda-eager

70/73/1

24758942.0

24401604.0

0.9855673154369844

0.907442229223269

1.0391748021708436

supported_no_material_regression

23673687.5 | 23434298.0 | 23185908.5 | 24760737.5 | 26693389.0 | 27603996.5 | 25135056.0 | 26533383.0

22850056.0 | 23580610.0 | 23888508.0 | 23211467.0 | 25226016.0 | 25727579.5 | 25695654.5 | 25433639.0

dense-256

cuda-cublas

67/70/1

9887812.0

9951557.0

1.006446825647575

0.8879440735084507

1.1246635587739058

inconclusive

8994399.5 | 9497337.5 | 9363270.0 | 9127860.0 | 12180325.5 | 10920574.5 | 10798179.5 | 10160297.0

9249476.0 | 9022598.0 | 8909743.5 | 9373977.5 | 11345134.0 | 11167892.0 | 11206536.0 | 11470192.0

dense-256

cuda-native

67/70/1

8178936.5

8184180.5

1.000641159153149

0.9029335389448141

1.1122471345705125

inconclusive

7545419.5 | 7924841.5 | 7548945.0 | 7618643.5 | 9263684.5 | 9396083.5 | 8784069.0 | 8614637.5

7717893.0 | 7655488.0 | 7427876.0 | 7939348.5 | 8616706.0 | 9170816.0 | 9981278.5 | 8921798.5

dense-256

torch-cuda-eager

67/70/1

16929456.5

17622993.5

1.0409662885515552

0.8995820777111698

1.1573567505752849

inconclusive

15553298.0 | 16267302.5 | 15477260.0 | 15571369.5 | 21288636.0 | 19434413.5 | 18343301.0 | 18288932.0

16137892.0 | 16124396.0 | 15178685.0 | 16785997.5 | 18436086.0 | 20199860.5 | 19395975.0 | 19164774.0

dense-64

cuda-cublas

49/51/0

7562041.0

7723447.0

1.0213442376205049

0.9361506611104952

1.0969273038243097

inconclusive

7009456.5 | 7180953.0 | 7305718.0 | 6950907.5 | 8517462.0 | 7867868.0 | 8029015.0 | 7766982.0

7157673.0 | 7015048.5 | 7112072.5 | 7286304.0 | 8089587.0 | 8104109.0 | 8173450.5 | 8157961.5

dense-64

cuda-native

49/51/0

5691617.0

5745350.0

1.0094407265984342

0.9197743695112001

1.1292273049107433

inconclusive

5290726.0 | 5541147.5 | 5266018.0 | 5226113.5 | 6332516.0 | 6175297.0 | 6173624.5 | 5930464.5

5341510.5 | 5381690.0 | 5211226.5 | 5636859.0 | 6249110.5 | 6218993.5 | 6269125.5 | 6417618.5

dense-64

torch-cuda-eager

49/51/0

12246492.5

12737790.5

1.0401174458727673

0.9188633831312408

1.1767315422076732

inconclusive

10997012.0 | 11813865.0 | 11276481.5 | 10997965.0 | 14056708.0 | 13971647.5 | 13466915.5 | 13707834.5

12046614.5 | 11674061.0 | 11071756.5 | 12374501.0 | 14358976.0 | 13688396.0 | 13998780.5 | 14542802.5

cuda-release-cuda-dense-f64-v1-none-api-fused0-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

70/73/1

13735302.5

13643809.5

0.993338843465588

0.9014193332338613

1.070136165583232

inconclusive

12702053.5 | 16161939.5 | 12768072.0 | 12677337.0 | 14233711.5 | 13542838.0 | 14590816.0 | 14101131.0

12476762.5 | 14240648.0 | 11833900.5 | 11956439.0 | 14300343.5 | 14533609.0 | 14286504.5 | 13671724.0

dense-1024

cuda-native

70/73/1

13533885.0

13547817.5

1.0010294531097317

0.8916026457187568

1.0721737816912944

inconclusive

12421125.0 | 15511000.5 | 12563515.5 | 12690692.0 | 14299236.5 | 13777114.5 | 14432435.0 | 13568292.5

12141497.0 | 14102276.0 | 11501194.5 | 11994740.0 | 13818864.5 | 14529511.0 | 13530732.5 | 13831015.5

dense-1024

torch-cuda-eager

70/73/1

19056846.0

18434373.0

0.9673359904361929

0.8584504246031479

1.0476226510937985

supported_no_material_regression

16335226.0 | 23193038.5 | 17970473.5 | 17567220.5 | 18434584.5 | 19158546.0 | 21439478.0 | 19031318.0

16529918.5 | 18759809.5 | 16039029.5 | 16319064.5 | 18976347.0 | 20194687.0 | 18932477.5 | 19193041.0

dense-2048

cuda-cublas

70/73/1

18888905.5

19464297.0

1.0304618761526443

0.9335368287955026

1.1408413079617263

inconclusive

17863783.5 | 16724740.5 | 17880647.5 | 17578923.0 | 19432164.0 | 19754248.0 | 20781868.5 | 20441380.0

17610193.0 | 17820905.5 | 23174901.0 | 17394568.0 | 20587745.0 | 19305193.0 | 22687248.5 | 19827768.5

dense-2048

cuda-native

70/73/1

18508620.0

19198717.5

1.0372851946822617

0.9381713003338719

1.1417440168604174

inconclusive

17538509.0 | 16365278.0 | 17383178.0 | 17222717.0 | 19064309.0 | 19059725.5 | 19966353.0 | 20174732.0

17043863.5 | 17431186.5 | 21642333.5 | 17517469.5 | 20198605.0 | 19231842.5 | 23668366.5 | 19627460.5

dense-2048

torch-cuda-eager

70/73/1

24239623.0

24797675.0

1.0230223052561502

0.9396435126799579

1.119683165591004

inconclusive

23212657.5 | 21910639.0 | 23258684.5 | 24167454.5 | 24185673.5 | 24557100.5 | 26412105.0 | 25877506.0

22929170.5 | 22651092.5 | 29711855.5 | 22422470.5 | 26491691.0 | 24708818.5 | 28112439.0 | 24913858.0

dense-256

cuda-cublas

67/70/1

10783777.5

10772900.5

0.9989913553019802

0.9259008061458489

1.0784313760318616

inconclusive

10136329.5 | 10197260.0 | 11206940.5 | 9786075.0 | 10741689.5 | 10773869.5 | 11680809.5 | 12096349.5

9592549.5 | 10102500.0 | 10765024.0 | 9624920.0 | 10909525.5 | 11550975.5 | 11241753.5 | 12203347.0

dense-256

cuda-native

67/70/1

9861600.5

9743450.5

0.9880191861351512

0.91185164690243

1.0893435739386217

inconclusive

9184265.5 | 9206631.0 | 10213852.5 | 9004786.0 | 9641578.5 | 9881391.5 | 10493160.5 | 10497707.5

8682173.0 | 9145435.5 | 9690647.0 | 8839163.0 | 10098762.5 | 10945901.0 | 10335410.5 | 11463815.5

dense-256

torch-cuda-eager

67/70/1

17847042.5

17928273.5

1.0045515104253269

0.8992213399036691

1.1052836712764327

inconclusive

16441141.0 | 16719854.0 | 19391575.5 | 16486982.5 | 17520627.0 | 18107156.5 | 19982198.5 | 18792032.5

15715078.5 | 16910234.5 | 17301059.5 | 16112437.0 | 18429166.0 | 19864343.0 | 19652423.0 | 20073203.5

dense-64

cuda-cublas

49/51/0

8052903.0

8567312.0

1.0638787031210981

0.9622635616558839

1.2421003784731077

inconclusive

7393973.5 | 7541461.5 | 7757618.0 | 7570049.0 | 8436634.5 | 8399139.5 | 9101639.5 | 8313834.0

7443926.5 | 7718352.0 | 11889208.0 | 7679041.0 | 8512798.5 | 10159427.0 | 8285391.5 | 10136919.5

dense-64

cuda-native

49/51/0

6371175.0

6890471.0

1.0815071003386345

0.9481611880520022

1.2531235314313096

inconclusive

5567345.0 | 5687028.5 | 6331217.0 | 5846424.5 | 6957502.5 | 6562191.0 | 7196656.5 | 6741021.0

5808394.0 | 6178699.5 | 8706502.5 | 5951849.0 | 6895969.0 | 7927625.5 | 6855592.0 | 8037933.0

dense-64

torch-cuda-eager

49/51/0

12655101.5

13835426.5

1.0932687106460584

0.9126335043887333

1.269592748858461

inconclusive

11234082.0 | 11164863.0 | 13574784.0 | 11492895.5 | 12820884.5 | 13270241.5 | 15380879.5 | 13718296.5

11119485.0 | 12046332.0 | 17467020.5 | 11463521.0 | 14204307.0 | 15665024.5 | 13361673.0 | 16088189.0

cuda-release-cuda-dense-f64-v1-none-api-fused1-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

70/73/1

9329260.5

9432042.0

1.0110171111633124

0.9432498537812176

1.077837354334355

inconclusive

8912039.0 | 9040014.0 | 8962627.5 | 8964262.5 | 10661362.0 | 9721854.5 | 9418422.0 | 9831650.5

8796779.5 | 8558905.0 | 9179902.5 | 8892046.5 | 10206417.5 | 10308602.0 | 9418331.5 | 9761605.0

dense-1024

cuda-native

70/73/1

8563303.5

8645569.5

1.009606806532082

0.9341406549391419

1.077433418402297

inconclusive

7862189.0 | 8233383.0 | 8406445.0 | 8137120.5 | 9713191.5 | 9310310.0 | 8384888.5 | 9042678.5

7766226.5 | 7561403.5 | 8889543.5 | 8087268.0 | 9352958.0 | 9188797.0 | 8633352.0 | 8783654.5

dense-1024

torch-cuda-eager

70/73/1

17677029.0

17746288.5

1.0039180509349168

0.92205622408761

1.0795857250625907

inconclusive

16961740.0 | 16817068.5 | 16964283.0 | 16433373.5 | 19547415.5 | 19514100.5 | 17970295.0 | 18331591.5

15673131.0 | 16326128.0 | 18266015.5 | 16827598.0 | 20203270.0 | 19040184.5 | 18131170.5 | 17806426.0

dense-2048

cuda-cublas

70/73/1

15958808.5

15792466.0

0.989576759442912

0.9276420813739705

1.0723088543822183

inconclusive

15182919.0 | 14955510.0 | 15229409.5 | 14962120.5 | 17195944.5 | 16135713.0 | 16375423.5 | 16760882.0

14881441.5 | 15212928.5 | 15163984.5 | 15085126.0 | 17221489.0 | 16841032.5 | 15989614.0 | 16853176.5

dense-2048

cuda-native

70/73/1

15184923.5

14984481.5

0.9867999334998296

0.9201705943262976

1.0928038131742306

inconclusive

14399054.5 | 14027106.0 | 14327838.0 | 14179046.0 | 16142969.0 | 15431250.5 | 15493376.0 | 15921095.0

14138464.5 | 14245753.5 | 14256804.0 | 14023060.0 | 16300362.5 | 16442490.0 | 15267594.0 | 16098558.0

dense-2048

torch-cuda-eager

70/73/1

24244496.5

23986288.5

0.9893498303831552

0.9199505182353187

1.0628305362979449

inconclusive

23403782.0 | 22453314.5 | 23707057.5 | 22493097.5 | 25880750.0 | 24650725.0 | 25622900.5 | 25701999.5

22435941.5 | 22259212.5 | 23311619.0 | 22321774.5 | 28110491.5 | 25495000.5 | 24612720.5 | 25834167.0

dense-256

cuda-cublas

67/70/1

7596811.5

7657468.5

1.007984534037734

0.9482664947860251

1.0882239876561814

inconclusive

7086665.5 | 6986157.5 | 7195676.5 | 7526087.5 | 8620340.5 | 7970554.5 | 7563982.5 | 7773724.5

7030129.5 | 7284451.5 | 7231876.5 | 7254034.5 | 8449503.5 | 8649880.0 | 7669956.0 | 7908964.0

dense-256

cuda-native

67/70/1

6084543.0

6005263.5

0.9869703443627565

0.9137773753684886

1.0704788466307815

inconclusive

5438653.5 | 5504736.0 | 5996946.0 | 6188348.0 | 6392744.0 | 6539582.0 | 5988389.0 | 6377390.5

5509551.5 | 5671850.0 | 5929646.0 | 5534103.5 | 6699757.0 | 6706806.5 | 6153238.0 | 6088102.5

dense-256

torch-cuda-eager

67/70/1

17517685.0

17507756.5

0.9994332299045222

0.9278545337774788

1.0884051038037197

inconclusive

15025741.0 | 16212264.0 | 16709208.5 | 17641670.5 | 19716423.5 | 19587389.5 | 17802512.5 | 17997048.5

15823303.5 | 16516352.5 | 17834632.5 | 16463423.0 | 19235723.0 | 19463895.5 | 17851629.5 | 17745613.0

dense-64

cuda-cublas

49/51/0

5916597.0

5896204.0

0.9965532551904414

0.9348539378757506

1.0568960167761994

inconclusive

5647137.5 | 5627129.5 | 5843689.5 | 5753545.5 | 6000650.5 | 6563966.0 | 6014910.0 | 6033745.5

5419295.5 | 5579840.0 | 5582271.0 | 5496277.0 | 6275129.5 | 6816842.5 | 5953914.5 | 6068351.0

dense-64

cuda-native

49/51/0

3977808.0

3933511.0

0.9888639672905278

0.9251353368072891

1.075294169244395

inconclusive

3635516.0 | 3650483.5 | 4016025.5 | 3779229.0 | 4137860.5 | 4755656.5 | 4002711.0 | 4126478.5

3701510.0 | 3927973.5 | 3730607.5 | 3654098.5 | 4250108.0 | 4581171.0 | 4063734.5 | 4060753.5

dense-64

torch-cuda-eager

49/51/0

12619252.0

12790344.5

1.0135580539955933

0.9269671707937984

1.1256645526579212

inconclusive

11131142.0 | 11786970.5 | 12975242.5 | 12289448.5 | 12464250.0 | 14759644.5 | 12701524.5 | 13433010.0

11409209.0 | 12862769.0 | 11852170.0 | 11996772.0 | 13984703.5 | 15558154.0 | 13390981.5 | 12949224.5

cuda-release-cuda-dense-f64-v1-none-api-fused0-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

70/73/1

9489961.5

9803104.0

1.0329972360794089

0.9505780618892239

1.1166253870459069

inconclusive

9040933.5 | 8660878.0 | 8551173.5 | 8746702.0 | 10601964.0 | 10036444.5 | 9843920.5 | 10851686.0

9932485.5 | 9537214.5 | 9107628.0 | 8720227.5 | 9939888.5 | 9953699.5 | 10185539.5 | 10249664.5

dense-1024

cuda-native

70/73/1

8789118.5

9098594.0

1.0352112103164839

0.948573738925486

1.1357318242614869

inconclusive

8316643.5 | 7899006.0 | 7925946.0 | 8033822.5 | 9751605.5 | 9223704.0 | 9218516.0 | 10055884.0

9174992.5 | 8784278.5 | 8516719.5 | 8017325.0 | 9081696.5 | 9447282.5 | 9457115.0 | 9752866.0

dense-1024

torch-cuda-eager

70/73/1

17630045.0

18064835.5

1.0246619052872525

0.9154722282025238

1.1684967273587394

inconclusive

16839678.5 | 15425104.5 | 15421181.0 | 15489690.5 | 20134264.0 | 18730477.5 | 18342372.0 | 20447438.5

18065602.5 | 16795338.0 | 17056198.0 | 15367298.0 | 18154960.5 | 18540288.0 | 19336212.0 | 19575030.5

dense-2048

cuda-cublas

70/73/1

16551380.5

16450083.0

0.9938798156443809

0.901921260424408

1.0751648078163938

inconclusive

15122538.5 | 14880179.0 | 16221994.5 | 16133903.0 | 17290390.0 | 17195841.5 | 16992951.5 | 17255721.0

14949560.0 | 14963954.0 | 16391069.5 | 15088569.0 | 16748711.5 | 17155261.5 | 16868535.0 | 17069761.0

dense-2048

cuda-native

70/73/1

15984230.5

15908426.0

0.9952575446156134

0.8968455169178945

1.062543818423581

inconclusive

14720702.0 | 14468301.5 | 15679976.5 | 15661348.5 | 16692592.0 | 16731083.5 | 16311263.0 | 16818142.0

14529043.5 | 14413729.0 | 15762477.0 | 14533980.0 | 16198341.5 | 16659161.5 | 16450371.0 | 16462636.0

dense-2048

torch-cuda-eager

70/73/1

24197094.5

24325871.5

1.0053220026065526

0.8837389931089424

1.0741246691693163

inconclusive

22753407.0 | 21597826.0 | 23339245.5 | 23117387.0 | 26339232.5 | 25820315.0 | 25344751.5 | 26159229.0

21705661.0 | 21481261.0 | 23816448.5 | 22192342.0 | 24821436.0 | 25901591.0 | 25180319.0 | 25668426.0

dense-256

cuda-cublas

67/70/1

8525381.5

8350054.0

0.9794346446549049

0.891957578631367

1.1219293468111038

inconclusive

7900721.5 | 7788281.0 | 7888315.5 | 7703697.5 | 8783635.5 | 9319209.0 | 9452460.0 | 9311300.5

7907816.5 | 7861570.5 | 7790725.5 | 7765000.0 | 8928886.0 | 9486691.0 | 9058431.5 | 9904500.0

dense-256

cuda-native

67/70/1

6850623.0

6743829.5

0.9844111258202356

0.8911835227828613

1.1095701936682274

inconclusive

6445178.5 | 6213929.0 | 6646877.0 | 6258651.0 | 7509381.5 | 7643847.0 | 7634915.5 | 7490384.5

6367881.5 | 6344494.5 | 6555423.0 | 6254954.5 | 7343771.0 | 7524750.0 | 7309176.0 | 8080172.0

dense-256

torch-cuda-eager

67/70/1

17848813.5

17333009.0

0.9711014684533512

0.8522072960223799

1.121020210862922

inconclusive

16520087.0 | 15400866.0 | 16823817.0 | 15447434.0 | 19180248.5 | 20066838.5 | 19027769.0 | 18723045.0

15443920.5 | 15436902.5 | 16250540.5 | 15385588.0 | 17947176.5 | 18870948.0 | 18794984.5 | 20221214.0

dense-64

cuda-cublas

49/51/0

6504231.5

6550975.0

1.0071866291967007

0.8889896867112622

1.1161907006937146

inconclusive

6075858.0 | 5968452.5 | 6097930.5 | 6010829.5 | 7757747.0 | 7221517.5 | 7119642.5 | 7370843.5

5995463.0 | 6131006.0 | 6123524.0 | 6085865.5 | 6857916.5 | 6882195.0 | 7698402.0 | 7623212.5

dense-64

cuda-native

49/51/0

4735092.0

4638227.0

0.9795431641032529

0.9118507184451413

1.0880270416882956

inconclusive

4617999.5 | 4242711.5 | 4732587.5 | 4280764.0 | 5430907.0 | 5038950.5 | 4847607.0 | 5122236.0

4394580.5 | 4382556.5 | 4573311.0 | 4344221.5 | 4956162.5 | 4880490.0 | 5613501.5 | 5015830.0

dense-64

torch-cuda-eager

49/51/0

12975718.5

12579421.0

0.9694585313329662

0.8483889416737554

1.120574970896572

inconclusive

12309431.0 | 10987048.0 | 12669888.5 | 11272065.0 | 14897653.0 | 13650248.5 | 13440921.5 | 14293826.0

11053130.5 | 11170253.0 | 11721504.0 | 11219909.0 | 13393906.0 | 13764168.5 | 14663377.5 | 14260529.5

cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

2943126.5

2964522.5

1.0072698200366175

0.9441052248675726

1.0845855832238418

inconclusive

2765305.0 | 2694964.5 | 2874390.5 | 2738497.0 | 3165219.5 | 3116311.5 | 3224581.5 | 3039135.0

2721490.0 | 3026472.5 | 2897319.0 | 2786042.5 | 2973869.5 | 3121448.5 | 2921582.0 | 3346772.5

dense-1024

cuda-native

4/6/0

1262342.0

1274488.5

1.0096221943023365

0.9532870195318506

1.080730191405936

inconclusive

1178474.0 | 1154600.5 | 1240907.0 | 1179193.5 | 1290828.5 | 1287811.5 | 1435261.5 | 1318544.0

1164282.0 | 1265579.0 | 1244221.5 | 1201880.0 | 1300261.0 | 1327631.5 | 1302155.0 | 1486936.5

dense-1024

torch-cuda-eager

4/6/0

2036332.5

2076437.5

1.0196947207786549

0.9493247977528319

1.1052043141342534

inconclusive

1953474.5 | 1845489.0 | 2013814.5 | 1914288.5 | 2129986.0 | 2129874.5 | 2300642.5 | 2141891.0

1909109.5 | 2050520.0 | 1973571.5 | 1949082.0 | 2127771.5 | 2266872.5 | 2029053.5 | 2364543.5

dense-2048

cuda-cublas

3/5/0

3202435.5

3098822.5

0.9676455622603484

0.9123139113705582

1.033652472117782

supported_no_material_regression

3006151.0 | 3016713.5 | 3134910.0 | 3013285.0 | 3279678.0 | 3363782.5 | 3323787.5 | 3465471.0

2998209.5 | 3169760.0 | 2976151.5 | 3074703.5 | 3088765.0 | 3161800.0 | 3354830.5 | 3452309.5

dense-2048

cuda-native

3/5/0

1582658.0

1560609.5

0.9860686895084093

0.9212493286459397

1.0422174051199413

supported_no_material_regression

1503626.5 | 1497388.5 | 1556005.0 | 1523805.5 | 1659037.5 | 1661709.5 | 1674105.0 | 1718526.0

1479613.0 | 1587795.5 | 1491953.0 | 1523445.5 | 1546143.0 | 1621608.0 | 1647874.5 | 1695662.5

dense-2048

torch-cuda-eager

3/5/0

2209522.0

2180053.0

0.9866627261461981

0.9109090372701051

1.0531018535572079

inconclusive

2065307.5 | 2090907.0 | 2375375.5 | 2073329.5 | 2367106.0 | 2331994.5 | 2299634.5 | 2411962.5

2068201.0 | 2172687.5 | 2038084.5 | 2127768.0 | 2157574.5 | 2276539.5 | 2321948.5 | 2471579.5

dense-256

cuda-cublas

4/6/0

2785805.5

2715144.5

0.9746353433504241

0.9204793794834467

1.0301971874055253

supported_no_material_regression

2567542.0 | 2493292.0 | 2722477.0 | 2623833.0 | 2960681.5 | 2879036.5 | 2866100.0 | 2843483.5

2528911.0 | 2746999.5 | 2519118.5 | 2552789.0 | 2798068.5 | 2894526.0 | 2740905.5 | 2865502.5

dense-256

cuda-native

4/6/0

966226.0

951230.5

0.9844803389683159

0.9358769471465642

1.0334720063559493

supported_no_material_regression

901962.0 | 885661.5 | 956451.0 | 923177.5 | 1012377.0 | 980730.0 | 1005628.0 | 986720.0

890142.5 | 962300.5 | 867172.0 | 905336.5 | 957077.5 | 1030716.5 | 986041.5 | 1043142.0

dense-256

torch-cuda-eager

4/6/0

1993731.0

1961271.0

0.9837189671023824

0.9259813541697579

1.0411661053804666

supported_no_material_regression

1924118.5 | 1840427.5 | 2000003.5 | 1974023.0 | 2022033.5 | 2209428.0 | 2126892.5 | 1978710.5

1872969.0 | 2049170.5 | 1872367.5 | 1890404.5 | 2023909.5 | 2225460.0 | 1955683.0 | 2056661.5

dense-64

cuda-cublas

5/7/0

2896332.5

2901369.0

1.0017389232762468

0.9454160917237727

1.0773050977438978

inconclusive

2685552.0 | 2636618.5 | 2835141.0 | 2657812.5 | 3126957.5 | 3239148.0 | 3033864.5 | 3000876.5

2812555.0 | 2874908.5 | 2724221.0 | 2654510.0 | 3069889.0 | 2980350.0 | 3037673.5 | 2974767.0

dense-64

cuda-native

5/7/0

1088990.0

1077152.5

0.9891298359029926

0.9320174485837207

1.0627406025910244

inconclusive

1001145.0 | 994099.5 | 1099716.5 | 993015.0 | 1124396.5 | 1155472.5 | 1131483.0 | 1178743.5

1041414.5 | 1071776.5 | 1008019.5 | 997920.0 | 1101030.5 | 1153981.0 | 1109522.5 | 1111076.5

dense-64

torch-cuda-eager

5/7/0

2284324.0

2309758.0

1.0111341473451227

0.9521438055450657

1.0732221150075372

inconclusive

2174554.0 | 2108923.0 | 2302708.0 | 2141449.0 | 2415914.5 | 2592603.0 | 2405068.5 | 2398999.5

2270223.5 | 2362980.0 | 2205097.5 | 2154432.0 | 2368521.0 | 2303802.0 | 2556480.0 | 2375836.5

cuda-release-cuda-dense-f64-v1-jacobi-api-fused0-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

3004754.5

2960002.5

0.9851062707452473

0.9414353782241234

1.0343385688792617

supported_no_material_regression

2927892.0 | 2788480.5 | 3088804.5 | 2809741.5 | 2988949.0 | 3035532.5 | 3056788.5 | 3097017.5

2854669.0 | 2784481.0 | 2787430.5 | 2928538.5 | 3000607.0 | 3054232.5 | 3116209.0 | 3128138.0

dense-1024

cuda-native

4/6/0

1357444.5

1333750.5

0.9825451427295923

0.9407040255706081

1.0263072357611738

supported_no_material_regression

1306937.5 | 1293998.0 | 1412627.5 | 1264890.5 | 1381744.5 | 1357430.5 | 1389525.0 | 1375839.0

1269521.0 | 1308532.5 | 1274323.0 | 1271431.5 | 1339705.0 | 1351542.0 | 1417102.5 | 1346818.0

dense-1024

torch-cuda-eager

4/6/0

2064730.5

2028662.5

0.9825313763709114

0.9246352582009699

1.0320889660369272

supported_no_material_regression

2041976.0 | 1893892.5 | 2140513.0 | 1957657.5 | 2104158.5 | 2106334.5 | 2275448.0 | 2064060.0

1923264.0 | 1944251.0 | 1868150.0 | 1921384.5 | 2154099.0 | 2026369.5 | 2262012.0 | 2084738.5

dense-2048

cuda-cublas

3/5/0

3178223.5

3152930.0

0.9920416232527385

0.9438726393614356

1.053823602506649

inconclusive

2938670.0 | 2972343.0 | 3289818.5 | 3044080.5 | 3286064.0 | 3380417.5 | 3196513.0 | 3204671.5

3058306.5 | 2962624.5 | 3100326.0 | 3015378.0 | 3298611.0 | 3239037.5 | 3385020.5 | 3287368.5

dense-2048

cuda-native

3/5/0

1638671.0

1610030.5

0.9825221170082341

0.9344856426667612

1.046223264574079

supported_no_material_regression

1539161.0 | 1561743.0 | 1715883.0 | 1554667.0 | 1697536.0 | 1765207.0 | 1627073.0 | 1619690.5

1567776.0 | 1534687.0 | 1530484.5 | 1538682.5 | 1699058.0 | 1665612.5 | 1729325.5 | 1664799.0

dense-2048

torch-cuda-eager

3/5/0

2228806.5

2202841.0

0.9883500429489953

0.9199475802057705

1.0604689675422814

inconclusive

2003113.0 | 2082136.0 | 2244112.0 | 2088116.5 | 2309751.0 | 2386441.5 | 2345092.5 | 2264487.0

2053562.0 | 2046867.5 | 2081120.0 | 2048948.5 | 2288735.0 | 2226619.5 | 2310004.5 | 2395540.5

dense-256

cuda-cublas

4/6/0

2732798.5

2768139.0

1.0129319816298201

0.9600858822363978

1.079920625451019

inconclusive

2535903.5 | 2620386.0 | 2596745.5 | 2562081.5 | 2797745.0 | 2793795.0 | 2844940.5 | 2867513.0

2792978.0 | 2664110.0 | 2592578.0 | 2538968.0 | 2793945.0 | 2774892.0 | 2926846.0 | 2906705.5

dense-256

cuda-native

4/6/0

1039283.0

1053243.0

1.013432337486517

0.9614286098629397

1.0697456395086284

inconclusive

978404.0 | 992581.0 | 1030401.0 | 970672.0 | 1071302.0 | 1052904.0 | 1121326.0 | 1072580.0

1050624.0 | 1010890.0 | 994446.0 | 967630.5 | 1066093.5 | 1069791.5 | 1091977.0 | 1139255.5

dense-256

torch-cuda-eager

4/6/0

1915379.0

1954705.5

1.0205319678246447

0.960630841420473

1.0744490806690554

inconclusive

1810818.5 | 1899858.5 | 1937311.0 | 1804071.5 | 1957037.5 | 2002590.5 | 2020512.5 | 1954849.5

2008875.5 | 1935846.5 | 1837902.0 | 1742760.0 | 2022495.5 | 2001254.5 | 2027424.5 | 2032564.5

dense-64

cuda-cublas

5/7/0

2908334.5

2943102.5

1.0119546083849709

0.9538454842176645

1.0738541164196973

inconclusive

2627829.5 | 2824362.5 | 2767920.0 | 2753585.0 | 3057625.5 | 2964584.5 | 3070957.0 | 3042455.0

2914073.0 | 2849128.0 | 2782002.0 | 2692467.5 | 2960764.5 | 2952742.0 | 3091719.0 | 3289209.5

dense-64

cuda-native

5/7/0

1136548.0

1153442.5

1.0148647483432287

0.9636490427596005

1.0739333430207885

inconclusive

1065433.5 | 1092876.0 | 1080169.5 | 1053901.5 | 1158294.0 | 1163422.0 | 1191401.5 | 1158475.0

1109346.0 | 1106515.0 | 1103340.5 | 1043184.5 | 1164134.0 | 1169008.5 | 1229449.0 | 1263287.0

dense-64

torch-cuda-eager

5/7/0

2261119.5

2326948.0

1.0291132335110993

0.9383465228129786

1.1232537272891006

inconclusive

2146374.5 | 2136976.5 | 2099596.0 | 2149135.5 | 2416117.5 | 2388893.0 | 2614047.5 | 2442303.0

2212384.0 | 2274935.0 | 2136338.5 | 2092704.5 | 2424372.0 | 2485523.5 | 2451291.5 | 2921849.5

cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

2688877.0

2648278.0

0.9849011315876479

0.9214712201810558

1.0257482098740867

supported_no_material_regression

2713117.5 | 2701681.5 | 2490884.0 | 2529029.5 | 2707109.5 | 2777167.0 | 2780583.5 | 2753667.5

2542770.5 | 2425567.5 | 2484170.0 | 2520886.5 | 2732377.0 | 2766514.5 | 2760232.0 | 2762914.0

dense-1024

cuda-native

4/6/0

1035070.5

1007491.0

0.9733549550489556

0.919655370018307

1.0220875280773059

supported_no_material_regression

1035455.5 | 1030689.5 | 944904.5 | 961745.5 | 1038293.5 | 1083440.5 | 1062237.0 | 1054994.5

957068.0 | 944791.5 | 936465.0 | 959833.5 | 1028746.0 | 1074132.5 | 1050067.0 | 1025239.5

dense-1024

torch-cuda-eager

4/6/0

2042824.5

1969252.5

0.963985158783831

0.9048361513555825

1.0124980173142257

supported_no_material_regression

2022068.0 | 2093950.5 | 1933829.0 | 1864358.0 | 2060695.5 | 2106422.5 | 2147686.5 | 2101795.5

1880122.0 | 1811547.5 | 1882692.0 | 1845695.5 | 2059742.5 | 2134747.0 | 2067463.0 | 2038823.5

dense-2048

cuda-cublas

3/5/0

3002448.5

2953550.5

0.9837139587906337

0.9161583913191107

1.0533344005429437

inconclusive

3167708.5 | 2752031.5 | 2830494.0 | 2901650.5 | 3012211.5 | 3087238.0 | 3102900.5 | 3169016.5

2799745.0 | 2632771.5 | 2972337.0 | 2805862.0 | 3056895.5 | 3349916.5 | 2962722.5 | 3006908.5

dense-2048

cuda-native

3/5/0

1427842.0

1393515.5

0.9759591747546297

0.9288516129514985

1.0440862091262408

supported_no_material_regression

1437217.0 | 1347915.0 | 1352291.0 | 1379329.5 | 1469233.5 | 1458067.0 | 1479631.5 | 1488774.5

1307929.0 | 1277793.5 | 1407333.5 | 1342577.0 | 1449818.0 | 1534713.5 | 1398561.0 | 1413283.0

dense-2048

torch-cuda-eager

3/5/0

2227446.0

2196441.5

0.986080695110005

0.9313647634999619

1.0433668517194141

supported_no_material_regression

2283441.5 | 2043720.5 | 2106382.0 | 2149351.5 | 2232109.0 | 2262883.0 | 2411192.0 | 2287763.0

2116787.0 | 1970817.5 | 2184521.5 | 2075106.5 | 2248061.0 | 2385638.5 | 2187393.0 | 2260343.0

dense-256

cuda-cublas

4/6/0

2519270.5

2496337.0

0.9908967695211769

0.932086481730748

1.043450740499508

supported_no_material_regression

2341803.0 | 2503312.0 | 2330202.5 | 2509244.5 | 2579710.5 | 2601959.5 | 2588312.0 | 2599991.0

2374551.0 | 2269326.5 | 2325007.5 | 2436398.5 | 2613364.5 | 2594636.0 | 2571094.0 | 2593703.0

dense-256

cuda-native

4/6/0

721506.0

709054.5

0.9827423472569875

0.923543487378433

1.043825744379471

supported_no_material_regression

668577.0 | 707165.5 | 656149.5 | 723277.5 | 742251.0 | 751169.0 | 730925.0 | 765782.0

672863.5 | 665585.5 | 662848.0 | 679740.0 | 742397.5 | 745087.5 | 740739.0 | 722576.0

dense-256

torch-cuda-eager

4/6/0

1941371.5

1922016.5

0.9900302440825983

0.9232988550775787

1.0514960976839534

inconclusive

1850909.5 | 1895158.0 | 1841243.5 | 1910836.5 | 1942134.0 | 2052872.5 | 2180453.0 | 2050842.0

1824243.0 | 1722632.5 | 1806610.5 | 1936814.5 | 1992507.5 | 2029846.0 | 2055793.5 | 2026052.5

dense-64

cuda-cublas

5/7/0

2555954.5

2559392.5

1.001345094366899

0.9291713177738042

1.0814479774536774

inconclusive

2328720.5 | 2431141.5 | 2397013.5 | 2416206.5 | 2608539.0 | 2692153.0 | 2705256.0 | 2739968.0

2373336.5 | 2339738.5 | 2415298.5 | 2464168.0 | 2646189.5 | 2743247.5 | 2711684.0 | 2669793.5

dense-64

cuda-native

5/7/0

751492.5

758032.5

1.0087026816634896

0.9424026950377741

1.0773588312085107

inconclusive

695074.5 | 710974.5 | 708271.5 | 697293.0 | 764917.5 | 788188.0 | 793951.5 | 807246.0

701565.5 | 708642.5 | 686476.5 | 719250.0 | 778524.0 | 787907.5 | 810986.5 | 773443.0

dense-64

torch-cuda-eager

5/7/0

2230315.0

2231673.0

1.000608882601785

0.9268126963141201

1.0744577508704116

inconclusive

2023419.0 | 2095012.0 | 2073035.0 | 2091325.0 | 2368843.0 | 2338249.5 | 2489489.0 | 2403927.5

2087273.0 | 2082761.0 | 2116868.5 | 2188932.0 | 2289207.5 | 2410489.5 | 2343725.5 | 2316978.5

cuda-release-cuda-dense-f64-v1-jacobi-api-fused0-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

2707830.5

2800627.0

1.0342696856394815

0.9392859733549515

1.115724577589087

inconclusive

2545634.0 | 2519776.0 | 2568433.5 | 2496469.5 | 3042895.5 | 2991552.5 | 2745868.0 | 2843415.0

2691989.0 | 2529750.5 | 2517100.5 | 2693237.5 | 2994877.0 | 2958007.0 | 2938982.0 | 2724219.5

dense-1024

cuda-native

4/6/0

1094311.5

1150974.0

1.0517791323585652

0.9171246224478539

1.1582810010221474

inconclusive

1016348.0 | 1027266.0 | 1033658.5 | 1007428.5 | 1260602.5 | 1239043.0 | 1115698.0 | 1104949.0

1132089.5 | 1017388.0 | 1026899.5 | 1077726.0 | 1259635.5 | 1246152.5 | 1216594.5 | 1125710.5

dense-1024

torch-cuda-eager

4/6/0

1975759.0

2012176.5

1.0184321569584145

0.9496918107034936

1.0907713148842568

inconclusive

1889037.0 | 1868645.5 | 1821927.0 | 1867743.0 | 2139183.5 | 2215853.5 | 2016566.0 | 2060137.5

1934208.0 | 1868815.0 | 1863704.0 | 2027968.5 | 2171927.0 | 2169743.5 | 2142212.5 | 2077501.0

dense-2048

cuda-cublas

3/5/0

2974604.5

2957337.0

0.9941950265993345

0.9355121771283066

1.058119418977182

inconclusive

2722540.0 | 3127448.0 | 2867710.0 | 2706598.5 | 3094490.5 | 3063392.0 | 3085095.0 | 3032953.5

3083765.0 | 2694638.5 | 2684081.0 | 2930526.5 | 3071488.0 | 3106945.5 | 3029146.5 | 2897872.0

dense-2048

cuda-native

3/5/0

1468881.0

1429138.0

0.9729433493931775

0.921201641031639

1.059028878645177

inconclusive

1343473.5 | 1551871.0 | 1405459.5 | 1319809.0 | 1506188.5 | 1503168.0 | 1532619.0 | 1496921.0

1498188.0 | 1336663.5 | 1339893.0 | 1422013.5 | 1502943.0 | 1558991.0 | 1456754.0 | 1384439.5

dense-2048

torch-cuda-eager

3/5/0

2139498.5

2132837.0

0.9968864198783033

0.936645319813995

1.0808954295455968

inconclusive

1959015.5 | 2206613.5 | 2043139.0 | 1987940.5 | 2268412.5 | 2230851.5 | 2249004.0 | 2251948.5

2131823.0 | 1977949.5 | 1975043.0 | 2129101.0 | 2308559.0 | 2240440.0 | 2301636.0 | 2255538.5

dense-256

cuda-cublas

4/6/0

2579571.5

2611139.5

1.0122376914150277

0.9312243450503759

1.081015855508905

inconclusive

2448444.5 | 2483192.0 | 2408152.5 | 2399768.0 | 2795522.5 | 2677514.0 | 2617458.5 | 2646690.0

2545783.0 | 2373453.0 | 2379789.0 | 2443969.5 | 2825149.0 | 2798369.0 | 2703890.0 | 2642463.0

dense-256

cuda-native

4/6/0

839433.5

860117.5

1.0246404271452116

0.8719135135165063

1.1403510395348255

inconclusive

797027.0 | 788220.0 | 770572.5 | 749087.0 | 981699.0 | 933189.5 | 826789.0 | 890791.0

856518.5 | 743227.0 | 743473.5 | 750250.0 | 962831.0 | 951375.0 | 942604.5 | 844333.5

dense-256

torch-cuda-eager

4/6/0

1929322.0

1938510.0

1.0047622947335904

0.9140752797360227

1.0977411603847176

inconclusive

1847883.0 | 1791894.5 | 1781020.5 | 1791268.0 | 2251955.0 | 2046294.0 | 2019449.0 | 2013294.0

1976029.5 | 1733309.5 | 1777275.5 | 1809233.5 | 2109833.5 | 2123981.0 | 2072886.0 | 2026840.5

dense-64

cuda-cublas

5/7/0

2666957.5

2752375.5

1.0320282569182297

0.9328288596035312

1.1044333694243411

inconclusive

2572155.5 | 2586378.0 | 2420140.0 | 2524705.5 | 2934807.5 | 2897203.0 | 2727856.0 | 2728916.0

2543374.0 | 2437385.0 | 2413015.5 | 2638804.5 | 2966373.0 | 2905225.0 | 2914223.5 | 2760556.5

dense-64

cuda-native

5/7/0

920527.0

954045.5

1.0364122942618739

0.907722820070633

1.130316374160149

inconclusive

877519.0 | 889683.5 | 827409.5 | 848737.0 | 1055852.0 | 1017416.0 | 917375.5 | 937624.5

872756.0 | 823547.0 | 823722.0 | 904315.0 | 1025888.0 | 1032133.0 | 1021914.0 | 964428.0

dense-64

torch-cuda-eager

5/7/0

2244040.5

2300677.0

1.025238626486465

0.9133480657456701

1.1048233274776615

inconclusive

2162062.5 | 2155732.0 | 2049816.0 | 2118189.0 | 2577678.5 | 2515425.0 | 2239308.5 | 2316748.0

2142338.0 | 2021798.5 | 2101211.5 | 2201902.5 | 2483328.5 | 2508159.5 | 2441341.0 | 2361299.5

cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

1216190.0

1201509.5

0.9879291064718506

0.9212576097894745

1.0576026965525658

inconclusive

1145861.5 | 1116390.0 | 1183854.5 | 1163089.0 | 1270782.5 | 1266975.0 | 1240254.5 | 1258134.0

1155223.5 | 1109858.5 | 1091640.5 | 1144501.0 | 1275612.0 | 1361541.5 | 1251838.5 | 1252248.5

dense-1024

cuda-native

4/6/0

1328617.0

1322090.0

0.9950873728094703

0.9233336885816879

1.0630339596536267

inconclusive

1260747.0 | 1235971.0 | 1251149.5 | 1257939.0 | 1391762.5 | 1384256.0 | 1365805.5 | 1390721.0

1327323.5 | 1198739.5 | 1206247.5 | 1221715.5 | 1381028.5 | 1452095.0 | 1370045.0 | 1362916.0

dense-1024

torch-cuda-eager

4/6/0

2201493.0

2193512.0

0.9963747329653103

0.9165533361856791

1.0848028482973708

inconclusive

2235272.5 | 2010571.0 | 2141320.5 | 2059067.5 | 2280676.0 | 2319658.0 | 2283779.0 | 2261332.5

2289571.5 | 1964918.5 | 1945282.5 | 2056132.5 | 2290823.5 | 2746140.5 | 2294578.5 | 2258247.0

dense-2048

cuda-cublas

3/5/0

1467639.0

1478018.5

1.0070722432423778

0.944975893773333

1.0918758835108908

inconclusive

1404191.5 | 1338896.5 | 1374288.0 | 1387550.5 | 1515220.5 | 1527668.5 | 1526933.5 | 1531994.0

1492428.0 | 1385101.0 | 1357717.5 | 1399074.5 | 1563283.5 | 1651175.0 | 1518327.5 | 1508788.5

dense-2048

cuda-native

3/5/0

1625734.5

1650519.5

1.0152454167639304

0.942668584268257

1.0882222123318952

inconclusive

1584086.0 | 1481659.5 | 1532264.0 | 1567745.5 | 1700510.5 | 1679999.5 | 1701616.5 | 1701588.5

1671059.5 | 1547238.5 | 1519294.0 | 1539231.5 | 1710202.5 | 1832309.5 | 1690361.0 | 1688922.5

dense-2048

torch-cuda-eager

3/5/0

2291808.5

2304215.5

1.0054136285819693

0.930610382676741

1.0949674553773454

inconclusive

2265174.0 | 2009660.0 | 2147290.0 | 2221325.5 | 2471700.0 | 2474528.5 | 2435297.0 | 2465247.0

2390134.5 | 2154307.5 | 2159912.5 | 2106722.5 | 2414554.5 | 2755554.5 | 2346757.5 | 2347950.5

dense-256

cuda-cublas

4/6/0

1007315.0

1020433.0

1.0130227386666535

0.9296016785599391

1.0785155225342915

inconclusive

964378.0 | 945642.5 | 970100.5 | 959106.0 | 1044934.0 | 1060791.0 | 1052441.5 | 1071638.0

1065639.5 | 908506.5 | 890199.5 | 974895.5 | 1071510.0 | 1169835.5 | 1053219.0 | 1037695.0

dense-256

cuda-native

4/6/0

1012511.0

1008746.0

0.9962815218797623

0.9271331478938312

1.0569116693587182

inconclusive

980315.0 | 951204.0 | 974677.5 | 949561.5 | 1069479.0 | 1053749.5 | 1039339.5 | 1073206.0

1025859.5 | 904421.0 | 933718.5 | 957701.5 | 1063736.5 | 1113576.5 | 1045884.0 | 1058148.0

dense-256

torch-cuda-eager

4/6/0

2183360.5

2191189.5

1.0035857569100477

0.924894893816511

1.0921956073651002

inconclusive

2132506.0 | 1984282.0 | 2089610.5 | 2077911.0 | 2275950.0 | 2294147.0 | 2278848.0 | 2319607.0

2267294.5 | 1929851.0 | 1967849.5 | 2071061.5 | 2292481.0 | 2577649.5 | 2279999.0 | 2247655.0

dense-64

cuda-cublas

5/7/0

1174507.5

1165134.5

0.9920196337613851

0.9044736681654356

1.1020287904742256

inconclusive

1199558.0 | 1049108.5 | 1044565.5 | 1075325.0 | 1197634.0 | 1229508.5 | 1240499.0 | 1220985.5

1145868.5 | 1042304.5 | 1057204.0 | 1055243.5 | 1250063.0 | 1339819.5 | 1207465.0 | 1248815.5

dense-64

cuda-native

5/7/0

1137889.0

1139802.5

1.001681622724185

0.9171753695373741

1.0996016006244236

inconclusive

1138093.5 | 1030085.0 | 995028.0 | 1066571.5 | 1177834.0 | 1224954.0 | 1176258.5 | 1200248.5

1135664.5 | 1022359.0 | 1044503.5 | 1036959.5 | 1157384.0 | 1316860.0 | 1192003.0 | 1207857.0

dense-64

torch-cuda-eager

5/7/0

2564735.5

2560390.5

0.9983058681879672

0.9137896896929124

1.1083702696788988

inconclusive

2715292.0 | 2250513.0 | 2325551.5 | 2295085.0 | 2657085.0 | 2594900.5 | 2745952.0 | 2708116.0

2618450.0 | 2267223.0 | 2304150.5 | 2325153.0 | 2660303.5 | 3182318.5 | 2628249.5 | 2688464.0

cuda-release-cuda-dense-f64-v1-jacobi-session-fused0-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

1300711.5

1282890.0

0.9862986526989267

0.9052868302387791

1.074891371814798

inconclusive

1171972.0 | 1187833.5 | 1384313.0 | 1135693.5 | 1488993.5 | 1324874.0 | 1318991.5 | 1372343.0

1256563.5 | 1233918.5 | 1176142.5 | 1134531.0 | 1316610.0 | 1318457.0 | 1319212.0 | 1344199.0

dense-1024

cuda-native

4/6/0

1450966.5

1412991.5

0.9738277899593133

0.9082747148256638

1.0739138132039512

inconclusive

1331656.0 | 1308968.0 | 1511737.5 | 1255564.5 | 1591421.5 | 1504715.0 | 1496821.0 | 1531536.5

1378516.0 | 1337630.0 | 1329845.5 | 1276790.5 | 1452505.5 | 1468064.0 | 1510701.0 | 1482976.5

dense-1024

torch-cuda-eager

4/6/0

2220417.5

2199740.0

0.9906875621364

0.8981725105520024

1.0886241867492839

inconclusive

2043296.5 | 1992080.0 | 2498687.5 | 1937957.5 | 2931122.5 | 2225291.0 | 2300844.5 | 2367325.5

2206187.0 | 2202474.0 | 2005825.5 | 1991421.0 | 2371309.0 | 2273322.0 | 2285596.5 | 2243313.5

dense-2048

cuda-cublas

3/5/0

1562149.5

1542564.5

0.9874627876525263

0.9294914682685661

1.059522997715453

inconclusive

1431171.5 | 1476623.0 | 1554909.5 | 1402246.0 | 1740997.5 | 1569980.5 | 1611958.0 | 1653029.0

1491190.0 | 1544549.5 | 1474586.0 | 1419372.0 | 1557760.5 | 1593361.5 | 1575797.0 | 1647847.0

dense-2048

cuda-native

3/5/0

1731466.5

1712362.0

0.9889662895585909

0.9329728146068759

1.0568227971195159

inconclusive

1605281.5 | 1682116.0 | 1720753.5 | 1593936.5 | 1886820.5 | 1728176.0 | 1813232.0 | 1823530.5

1672476.0 | 1690570.5 | 1642082.0 | 1566000.0 | 1751177.5 | 1748078.0 | 1797487.0 | 1798665.5

dense-2048

torch-cuda-eager

3/5/0

2365782.0

2349356.5

0.993057052594026

0.9303796785131906

1.0586374920456687

inconclusive

2186984.5 | 2289968.5 | 2382992.0 | 2114600.0 | 2815254.5 | 2383869.0 | 2511327.0 | 2460878.0

2290365.5 | 2425135.0 | 2276189.5 | 2091357.0 | 2347943.0 | 2470869.0 | 2416470.5 | 2482565.0

dense-256

cuda-cublas

4/6/0

1073989.0

1079391.5

1.0050303122285238

0.9271370187451898

1.101686255480773

inconclusive

975664.5 | 977445.0 | 1080596.0 | 912030.5 | 1212427.0 | 1142148.5 | 1135625.5 | 1127206.5

1031113.5 | 1053000.5 | 989866.5 | 956522.0 | 1115437.0 | 1104038.0 | 1131788.5 | 1144358.5

dense-256

cuda-native

4/6/0

1113693.0

1110659.5

0.9972761793420628

0.9268616342653282

1.0779837698959989

inconclusive

1051133.5 | 1021747.0 | 1134583.0 | 948434.0 | 1254581.5 | 1155594.5 | 1173221.0 | 1170129.5

1072298.5 | 1068550.0 | 1030186.0 | 1017622.5 | 1141465.5 | 1140507.0 | 1176660.0 | 1158285.0

dense-256

torch-cuda-eager

4/6/0

2156103.0

2180696.5

1.0114064587823495

0.9156484013430063

1.1276274701277553

inconclusive

1949026.0 | 1935152.5 | 2191602.0 | 1794611.5 | 2627777.0 | 2309541.5 | 2251096.5 | 2289862.0

2097322.0 | 2219392.0 | 1975530.0 | 1955988.5 | 2298768.5 | 2261713.0 | 2273320.5 | 2249602.5

dense-64

cuda-cublas

5/7/0

1246988.0

1224450.5

0.981926449973857

0.8744960884248144

1.0772822033165925

inconclusive

1130458.0 | 1149230.5 | 1157895.0 | 1139463.5 | 1456948.0 | 1318710.0 | 1344352.5 | 1343792.0

1202444.0 | 1136169.0 | 1096984.5 | 1106576.0 | 1285869.0 | 1309526.5 | 1292654.0 | 1272900.5

dense-64

cuda-native

5/7/0

1226591.5

1207992.5

0.9848368425836964

0.8941607799658271

1.0854503643504836

inconclusive

1121752.5 | 1125522.0 | 1157445.5 | 1137681.5 | 1365644.0 | 1279989.5 | 1331862.5 | 1291282.0

1187962.5 | 1129462.5 | 1076852.0 | 1109163.5 | 1256310.5 | 1450742.0 | 1326314.0 | 1291895.5

dense-64

torch-cuda-eager

5/7/0

2486444.5

2484477.5

0.9992089105548103

0.8835617069489489

1.0927788345400005

inconclusive

2244687.5 | 2307693.0 | 2339831.5 | 2352364.5 | 3102649.5 | 2662937.0 | 2818089.0 | 2752412.5

2547510.5 | 2333706.0 | 2217205.0 | 2204564.5 | 2740713.5 | 2619117.5 | 2569293.0 | 2578637.0

cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

929326.0

941544.5

1.0131477005916116

0.9303165175423873

1.088587911839952

inconclusive

842059.0 | 865308.5 | 931433.0 | 886359.5 | 952263.0 | 983354.0 | 950808.5 | 964295.5

848267.0 | 847686.5 | 969330.0 | 874935.5 | 956142.0 | 983307.5 | 975625.5 | 988362.5

dense-1024

cuda-native

4/6/0

1045467.0

1057920.0

1.0119114233160875

0.9443350052279496

1.0842124650423752

inconclusive

941915.5 | 985939.0 | 1047999.0 | 1003079.5 | 1071626.0 | 1106445.0 | 1073015.5 | 1096223.5

974658.0 | 978909.5 | 1112063.0 | 1017994.5 | 1087415.5 | 1109317.0 | 1096138.0 | 1087860.5

dense-1024

torch-cuda-eager

4/6/0

2200508.5

2196516.0

0.9981856466357663

0.9198376164162703

1.1176420743001636

inconclusive

1929310.0 | 2039979.0 | 2467275.0 | 1995897.0 | 2200508.5 | 2336204.5 | 2308356.5 | 2300552.5

1946037.5 | 1955296.0 | 2410985.0 | 2161973.0 | 2231229.5 | 2242564.5 | 2350350.0 | 2317267.5

dense-2048

cuda-cublas

3/5/0

1281745.0

1300345.5

1.0145118568826093

0.9472708862412362

1.0833575907495991

inconclusive

1175746.0 | 1173316.5 | 1268225.5 | 1245887.0 | 1330330.5 | 1323153.0 | 1351148.0 | 1349887.0

1194370.5 | 1176795.5 | 1239916.5 | 1303162.5 | 1331925.5 | 1366426.0 | 1345418.5 | 1333411.5

dense-2048

cuda-native

3/5/0

1453268.0

1492419.5

1.026940316583039

0.9467251562699242

1.0939296942594567

inconclusive

1338400.5 | 1360254.0 | 1448733.0 | 1424587.5 | 1483558.5 | 1511623.5 | 1527926.0 | 1493009.0

1337287.0 | 1339745.5 | 1441774.5 | 1481957.5 | 1510008.5 | 1597118.5 | 1539211.5 | 1537112.5

dense-2048

torch-cuda-eager

3/5/0

2297320.0

2352666.5

1.0240917677990007

0.9482416444202861

1.1021984822574564

inconclusive

2104769.0 | 2075653.5 | 2284737.0 | 2242746.5 | 2377070.5 | 2473984.0 | 2430972.0 | 2389962.0

2157297.5 | 2109897.5 | 2292062.5 | 2302102.5 | 2423254.0 | 2406136.5 | 2475667.5 | 2406266.0

dense-256

cuda-cublas

4/6/0

762010.0

794600.0

1.042768467605412

0.9506810572562605

1.1265469899190845

inconclusive

718682.5 | 701919.5 | 785023.5 | 705711.0 | 791826.5 | 819429.5 | 829274.5 | 790614.5

707384.5 | 699682.5 | 814767.0 | 815318.0 | 821471.0 | 829506.5 | 851607.0 | 850977.0

dense-256

cuda-native

4/6/0

767230.0

768009.0

1.0010153409016853

0.9439682443114128

1.0728772460314044

inconclusive

712690.5 | 720799.5 | 786727.5 | 707824.0 | 778094.0 | 828968.5 | 796728.0 | 783868.0

697438.0 | 702494.0 | 812625.5 | 744595.5 | 771225.0 | 824813.0 | 805406.0 | 778305.0

dense-256

torch-cuda-eager

4/6/0

2226846.0

2211156.5

0.9929543848115227

0.8990902822632922

1.0950532169594758

inconclusive

1993726.0 | 2025308.0 | 2540261.5 | 2005767.0 | 2293195.0 | 2365478.5 | 2331670.0 | 2265014.0

1955971.5 | 1928141.0 | 2341941.0 | 2262287.0 | 2198868.0 | 2428528.0 | 2238755.0 | 2265965.5

dense-64

cuda-cublas

5/7/0

848014.0

858000.0

1.0117757489852761

0.9389741257189883

1.1017997644324196

inconclusive

773833.0 | 744730.5 | 905286.5 | 803683.5 | 868190.0 | 890608.5 | 899553.0 | 884155.5

781755.5 | 788724.0 | 948562.0 | 775185.5 | 890889.5 | 928777.0 | 849979.5 | 887772.5

dense-64

cuda-native

5/7/0

789502.0

788972.0

0.999328690744292

0.9371807520600642

1.075870111964512

inconclusive

711627.0 | 713130.0 | 852885.0 | 758920.0 | 804605.5 | 855549.5 | 825740.5 | 812755.0

739905.0 | 743624.0 | 864554.5 | 745533.5 | 793091.5 | 859342.5 | 769896.5 | 829759.5

dense-64

torch-cuda-eager

5/7/0

2527079.0

2518408.0

0.9965687657568283

0.8968400590635929

1.1290818124588384

inconclusive

2225701.5 | 2163553.0 | 2804693.0 | 2239444.5 | 2597354.5 | 2719063.0 | 2605568.5 | 2595554.5

2333009.5 | 2180441.5 | 2806368.5 | 2382023.0 | 2634575.0 | 2709066.5 | 2342589.0 | 2672080.0

cuda-release-cuda-dense-f64-v1-jacobi-session-fused0-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

1067459.5

1125186.0

1.0540783982905206

0.9488132827014996

1.1882251089445044

inconclusive

1004947.0 | 916650.5 | 917429.0 | 944787.0 | 1153192.0 | 1154256.5 | 1213865.0 | 1191959.0

1229717.0 | 1087226.0 | 966065.0 | 1189174.0 | 1141001.5 | 1158886.0 | 1125893.5 | 1141641.5

dense-1024

cuda-native

4/6/0

1192792.0

1260056.5

1.0563924808348815

0.9669733931166589

1.183122041472135

inconclusive

1142861.0 | 1021939.0 | 1028545.0 | 1081508.5 | 1269811.5 | 1271329.0 | 1330369.5 | 1317334.5

1315503.5 | 1195407.0 | 1091277.0 | 1285926.0 | 1288462.5 | 1286635.5 | 1236347.5 | 1264767.5

dense-1024

torch-cuda-eager

4/6/0

2129443.5

2292739.5

1.0766848239927473

0.94896373519422

1.1971414181269395

inconclusive

2096268.5 | 1915779.5 | 1894120.0 | 2010379.0 | 2208152.5 | 2238315.5 | 2507974.5 | 2627448.5

2576123.0 | 2221322.0 | 1892133.0 | 2582117.5 | 2318562.5 | 2277860.0 | 2241093.0 | 2274547.0

dense-2048

cuda-cublas

3/5/0

1423191.0

1433000.5

1.0068926096356707

0.942037073942941

1.0617022274452128

inconclusive

1375780.5 | 1257734.5 | 1333845.5 | 1400753.5 | 1500640.5 | 1484166.0 | 1487064.0 | 1461197.5

1412188.0 | 1265905.5 | 1314997.5 | 1435815.5 | 1467705.5 | 1462858.0 | 1456642.5 | 1454712.5

dense-2048

cuda-native

3/5/0

1615090.5

1621847.0

1.0041833569078638

0.9388481935142954

1.0642129263664182

inconclusive

1566455.0 | 1409432.0 | 1515589.0 | 1581945.5 | 1672025.0 | 1686157.5 | 1663297.5 | 1652905.0

1606923.5 | 1425122.5 | 1471664.0 | 1652373.5 | 1661312.0 | 1659334.0 | 1647011.5 | 1638563.0

dense-2048

torch-cuda-eager

3/5/0

2342077.5

2418135.0

1.0324743737130817

0.95145926678245

1.0940780843323505

inconclusive

2300718.0 | 2125453.0 | 2226290.5 | 2244497.0 | 2508055.5 | 2518179.0 | 2563699.0 | 2488894.0

2560220.5 | 2089812.0 | 2273675.5 | 2563013.5 | 2511964.0 | 2408262.5 | 2435575.5 | 2421124.0

dense-256

cuda-cublas

4/6/0

913036.0

940534.0

1.0301171038162789

0.9644819451195389

1.1063369446208449

inconclusive

879565.0 | 839209.0 | 805611.5 | 913400.0 | 945561.0 | 943975.0 | 981853.5 | 1035216.0

936375.5 | 904418.5 | 819755.0 | 962551.5 | 983708.5 | 973452.0 | 935380.0 | 959851.5

dense-256

cuda-native

4/6/0

922383.5

961497.5

1.0424053552562464

0.9707764176248809

1.10971894732601

inconclusive

869639.5 | 865405.0 | 830071.0 | 901372.0 | 959135.0 | 965025.0 | 1011567.0 | 991363.0

980885.0 | 957233.5 | 826884.0 | 952766.5 | 999360.5 | 987243.5 | 945602.5 | 971088.0

dense-256

torch-cuda-eager

4/6/0

2136933.5

2290718.5

1.071965271731666

0.9483378674203049

1.1633206798765656

inconclusive

2130498.5 | 1918677.5 | 1901667.5 | 2039350.0 | 2273819.0 | 2331075.5 | 2551742.5 | 2630012.5

2490311.5 | 2362253.5 | 1883402.5 | 2378158.5 | 2329571.0 | 2297093.5 | 2260588.0 | 2278949.0

dense-64

cuda-cublas

5/7/0

997768.0

1055406.5

1.0577674369192036

0.977618925777081

1.1660871454948538

inconclusive

882595.5 | 887331.5 | 889124.0 | 948880.5 | 1051252.0 | 1034542.5 | 1079744.0 | 1143201.0

1177628.0 | 1022579.0 | 857828.0 | 1106809.5 | 1057579.0 | 1065791.5 | 1044174.0 | 1082682.5

dense-64

cuda-native

5/7/0

993369.0

1036355.5

1.0432734462218973

0.957803354990063

1.1775216358357465

inconclusive

891592.5 | 839363.5 | 859228.0 | 946919.0 | 1056070.5 | 1035518.5 | 1102871.5 | 1109925.0

1181446.0 | 937064.0 | 852226.0 | 1077955.5 | 1073075.5 | 1053979.5 | 1067165.5 | 1070043.0

dense-64

torch-cuda-eager

5/7/0

2430322.0

2688661.5

1.106298465800005

0.9743237675837257

1.201865151133442

inconclusive

2439679.5 | 2176711.0 | 2193797.0 | 2331236.0 | 2562470.0 | 2374895.0 | 2847888.5 | 3068987.0

2920560.5 | 2711385.0 | 2201456.5 | 2753745.0 | 2797227.0 | 2674370.0 | 2646163.5 | 2686265.5

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

1198724.0

1177950.0

0.9826699056663586

0.9245637702711879

1.0282447375324737

supported_no_material_regression

1201929.0 | 1165197.5 | 1316825.0 | 1372490.5 | 1192920.5 | 1178952.5 | 1186678.0 | 1152269.5

1096567.5 | 1090622.0 | 1293420.5 | 1116398.0 | 1163117.0 | 1169808.5 | 1176388.5 | 1240749.0

dense-1024

cuda-native

4/6/0

1315134.0

1275629.0

0.9699612358892706

0.905705007836203

1.0357388291239362

supported_no_material_regression

1391384.5 | 1263141.5 | 1491538.0 | 1497969.5 | 1287869.5 | 1269949.0 | 1304705.0 | 1277248.0

1198784.0 | 1177046.5 | 1507939.5 | 1242543.0 | 1250038.5 | 1262477.0 | 1278883.5 | 1322279.0

dense-1024

torch-cuda-eager

4/6/0

1950882.0

1857558.0

0.9521631754252692

0.8869846183488035

1.016088756031546

supported_no_material_regression

2093010.5 | 1834560.5 | 2163965.5 | 2118343.0 | 1910995.5 | 1887328.0 | 1947325.5 | 1884412.5

1786271.5 | 1682711.5 | 2150068.0 | 1855409.5 | 1805748.5 | 1837469.5 | 1827017.0 | 1976670.0

dense-2048

cuda-cublas

3/5/0

1495207.0

1444073.0

0.9658013907104501

0.9214857815982693

1.046353019385354

supported_no_material_regression

1463426.5 | 1532744.0 | 1669980.5 | 1479284.5 | 1445284.5 | 1427702.0 | 1471737.0 | 1420488.0

1407547.0 | 1330502.5 | 1791250.5 | 1488567.5 | 1421935.5 | 1420559.0 | 1443846.5 | 1505039.5

dense-2048

cuda-native

3/5/0

1645518.5

1621102.5

0.9851621236710496

0.9221568239829265

1.067246732391509

inconclusive

1568839.0 | 1721723.5 | 1962836.5 | 1658674.5 | 1610281.0 | 1613484.5 | 1607392.0 | 1603386.5

1566647.0 | 1502064.0 | 2016516.5 | 1665414.0 | 1613364.0 | 1596072.0 | 1630588.0 | 1684931.0

dense-2048

torch-cuda-eager

3/5/0

2064055.5

2027854.5

0.9824612274233905

0.9170160971101025

1.0454533903886558

supported_no_material_regression

2108488.0 | 2065683.0 | 2572512.5 | 2211354.0 | 1999783.5 | 1947713.5 | 2058002.5 | 1973149.0

1984461.0 | 1846732.5 | 2674419.0 | 2032669.5 | 2014494.5 | 2010791.5 | 1998154.5 | 2132914.5

dense-256

cuda-cublas

4/6/0

979547.0

966529.0

0.9867101833806852

0.9370118485154556

1.032704392822436

supported_no_material_regression

1048601.0 | 919613.0 | 968350.5 | 1173339.5 | 977426.5 | 953483.0 | 986231.0 | 977804.5

948308.5 | 954312.0 | 984945.5 | 874098.5 | 943505.5 | 997420.0 | 988276.5 | 994549.0

dense-256

cuda-native

4/6/0

960781.5

958936.5

0.9980796882537809

0.9560490568856199

1.0323775648534856

supported_no_material_regression

974276.0 | 916757.0 | 983448.0 | 1146199.0 | 968306.5 | 941323.0 | 941161.5 | 972891.5

926905.5 | 933535.5 | 980471.5 | 885573.0 | 942650.5 | 962101.0 | 983821.5 | 995036.5

dense-256

torch-cuda-eager

4/6/0

1879402.5

1889673.0

1.0054647687230382

0.9025863902795453

1.0622239994331375

inconclusive

2140810.5 | 1756231.5 | 1805175.0 | 2373295.0 | 1913466.5 | 1820473.5 | 1953080.0 | 1873051.5

1956433.0 | 1934070.5 | 1913120.5 | 1683256.0 | 1827119.5 | 1921345.0 | 1903571.5 | 1913437.5

dense-64

cuda-cublas

5/7/0

1132603.5

1119967.5

0.988843403715422

0.9326370694835772

1.0326967796670803

supported_no_material_regression

1213701.0 | 1089801.0 | 1221899.0 | 1189995.5 | 1143629.5 | 1058474.0 | 1129484.0 | 1094812.5

1116494.0 | 1029126.5 | 1142724.0 | 1074994.0 | 1103366.0 | 1135031.0 | 1142754.5 | 1150235.0

dense-64

cuda-native

5/7/0

1084753.0

1089246.5

1.004142417674807

0.9541571279945102

1.0470891566046454

supported_no_material_regression

1128895.5 | 1041483.0 | 1193190.0 | 1071019.5 | 1082272.5 | 1041859.0 | 1094032.5 | 1061385.5

1086816.0 | 993603.5 | 1080981.0 | 1072719.0 | 1089246.5 | 1087223.0 | 1153624.5 | 1097199.5

dense-64

torch-cuda-eager

5/7/0

2224705.0

2244476.0

1.0088870209758147

0.9002462079201982

1.063204220233008

inconclusive

2577281.0 | 2083809.5 | 2545141.0 | 2575451.0 | 2226055.0 | 2057893.5 | 2289738.0 | 2121290.5

2281922.5 | 1944427.5 | 2262208.0 | 2256639.0 | 2208850.5 | 2204295.0 | 2349825.5 | 2251717.5

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused0-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

1226955.0

1256219.0

1.0238509154777478

0.9897566268606006

1.1027697197056372

inconclusive

1465411.5 | 1191186.5 | 1183940.0 | 1141210.0 | 1239067.0 | 1243448.0 | 1256996.5 | 1232378.5

1462769.0 | 1414912.5 | 1246563.5 | 1253732.5 | 1223733.5 | 1224407.5 | 1292009.0 | 1230655.5

dense-1024

cuda-native

4/6/0

1363074.5

1380985.5

1.0131401475121133

0.9858878048320452

1.0523933374088903

inconclusive

1517901.0 | 1335030.5 | 1333836.5 | 1293450.5 | 1401658.0 | 1354036.5 | 1386414.0 | 1395311.5

1559388.5 | 1484925.0 | 1369160.5 | 1378145.0 | 1348585.5 | 1358718.5 | 1382279.5 | 1377612.5

dense-1024

torch-cuda-eager

4/6/0

1916840.5

1957552.5

1.0212391171826765

0.9459477223179246

1.145271475137628

inconclusive

2387748.0 | 1950638.5 | 1789069.0 | 1674960.5 | 1980889.5 | 1843893.5 | 1909640.5 | 1916083.0

2514356.5 | 2342666.0 | 1911426.0 | 2064321.0 | 1790915.5 | 1877469.0 | 1805181.5 | 1957761.0

dense-2048

cuda-cublas

3/5/0

1475034.0

1508463.5

1.0226635453826827

0.989242417808728

1.0722812814713893

inconclusive

1392715.0 | 1453536.5 | 1430278.0 | 1441080.0 | 1551279.0 | 1507026.5 | 1480388.5 | 1524086.0

1580653.5 | 1490099.0 | 1708198.0 | 1488773.0 | 1486458.0 | 1500592.5 | 1541444.0 | 1472247.5

dense-2048

cuda-native

3/5/0

1670465.5

1708558.0

1.0228035239279112

0.9805454229803171

1.068800634951112

inconclusive

1559577.5 | 1644998.0 | 1613556.5 | 1641687.5 | 1779343.0 | 1671194.0 | 1705944.0 | 1742755.0

1644638.5 | 1712470.5 | 1848620.0 | 1686256.5 | 1709011.5 | 1677779.5 | 1764336.5 | 1666664.0

dense-2048

torch-cuda-eager

3/5/0

1978325.5

2009721.5

1.0158699870167978

0.9678152766592781

1.0861255993960675

inconclusive

1800973.5 | 2020328.5 | 1845106.0 | 1942156.0 | 2146550.0 | 2070011.5 | 1936167.5 | 2040676.5

2090552.5 | 1972057.5 | 2364373.0 | 1995152.5 | 2002930.0 | 2006318.0 | 2073332.0 | 1941569.5

dense-256

cuda-cublas

4/6/0

1018567.5

1049699.5

1.0305644937620728

0.9655124401527617

1.10326479514228

inconclusive

938146.0 | 1015453.5 | 935514.5 | 975513.0 | 1108287.5 | 999788.0 | 1083649.5 | 1041041.0

1322100.5 | 1038837.0 | 970190.0 | 1178372.5 | 1077825.0 | 1005644.0 | 1041110.5 | 1046031.5

dense-256

cuda-native

4/6/0

1052439.5

1083742.5

1.0297432774045443

0.9735311643413389

1.0847279070161018

inconclusive

1021406.0 | 1047884.5 | 987785.0 | 1001173.5 | 1129914.5 | 1018776.5 | 1121775.0 | 1097291.5

1323019.0 | 1078370.0 | 1011039.5 | 1094337.0 | 1122850.5 | 1069224.5 | 1092622.5 | 1078045.0

dense-256

torch-cuda-eager

4/6/0

1878372.5

1919502.5

1.0218966152879687

0.9411720002797029

1.1282517629370221

inconclusive

1751599.0 | 2006926.0 | 1705593.5 | 1710776.0 | 2091094.5 | 1772730.0 | 1967792.0 | 1916245.5

2359060.0 | 1919502.5 | 1696639.0 | 2100690.5 | 2120198.0 | 1875784.0 | 1875589.0 | 1864689.0

dense-64

cuda-cublas

5/7/0

1187241.5

1255696.5

1.0576588672144631

0.9922935932017973

1.1363437036121757

inconclusive

1119651.5 | 1186863.0 | 1085130.0 | 1141106.5 | 1304362.0 | 1225980.5 | 1274164.0 | 1203314.5

1546083.0 | 1232703.0 | 1359377.0 | 1288218.5 | 1281046.0 | 1182147.5 | 1213607.0 | 1241222.5

dense-64

cuda-native

5/7/0

1191263.0

1258162.0

1.0561580440255427

1.0035125667160165

1.1419745924646485

inconclusive

1102446.5 | 1188083.5 | 1090439.0 | 1128958.0 | 1311350.0 | 1202673.0 | 1223840.0 | 1227773.5

1559609.0 | 1307700.0 | 1247306.0 | 1237880.5 | 1316490.5 | 1190494.5 | 1254996.0 | 1215431.5

dense-64

torch-cuda-eager

5/7/0

2221459.5

2354417.5

1.0598516425800246

0.97570452171871

1.1849822466377764

inconclusive

2075051.0 | 2311725.5 | 1969136.5 | 1991646.0 | 2598687.5 | 2269641.5 | 2394123.5 | 2313698.5

2852004.0 | 2551085.0 | 2659063.0 | 2291604.0 | 2395188.0 | 2241673.5 | 2205059.5 | 2226508.0

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

879614.0

870428.0

0.9895567828615733

0.9629340623484879

1.015040417146673

supported_no_material_regression

879410.0 | 834461.0 | 911731.0 | 893130.0 | 856309.0 | 878079.5 | 886903.5 | 893522.5

851249.0 | 858628.5 | 846059.0 | 847181.5 | 898335.0 | 874490.0 | 887191.5 | 879738.5

dense-1024

cuda-native

4/6/0

1003462.0

1006485.5

1.0030130687559669

0.9792696920566465

1.0291229568046711

supported_no_material_regression

1056475.5 | 945434.5 | 1008428.5 | 984574.0 | 1001847.0 | 1017708.0 | 1001562.0 | 1018250.5

987138.0 | 1064440.5 | 975643.5 | 968265.5 | 1025571.0 | 1020306.5 | 1000104.0 | 1017447.5

dense-1024

torch-cuda-eager

4/6/0

1875338.5

1823466.5

0.9723399268985305

0.9297787156068065

1.0122080572177732

supported_no_material_regression

1904923.5 | 1778026.5 | 2037481.0 | 2048537.0 | 1848659.0 | 1896591.5 | 1812935.5 | 1833385.5

1805700.0 | 1862354.0 | 1747324.5 | 1857940.0 | 1915036.5 | 1829206.0 | 1820018.5 | 1842276.0

dense-2048

cuda-cublas

3/5/0

1243527.0

1234146.5

0.9924565369308427

0.9665553526121174

1.021830878421905

supported_no_material_regression

1295880.5 | 1183883.0 | 1249486.0 | 1259549.0 | 1243952.0 | 1253789.5 | 1201750.5 | 1255422.5

1242012.0 | 1169721.5 | 1204863.5 | 1225521.0 | 1252299.0 | 1254120.5 | 1263008.5 | 1246713.0

dense-2048

cuda-native

3/5/0

1429591.0

1424490.5

0.9964321963414711

0.9718980901118215

1.0208057290887316

supported_no_material_regression

1464777.5 | 1366875.0 | 1431690.5 | 1446783.0 | 1418136.0 | 1445137.5 | 1382179.5 | 1441249.5

1426429.5 | 1367651.5 | 1389696.5 | 1444138.0 | 1436322.5 | 1450150.5 | 1443893.5 | 1418171.0

dense-2048

torch-cuda-eager

3/5/0

1996998.5

1977412.5

0.9901922810658095

0.949158049639772

1.0375139857747553

supported_no_material_regression

2023970.5 | 1869356.0 | 2032337.5 | 2150794.5 | 2022524.5 | 1999440.0 | 1849035.5 | 2000899.5

2018095.0 | 1875678.5 | 1911652.0 | 1989932.5 | 1966377.5 | 2118375.0 | 2024845.5 | 2009784.5

dense-256

cuda-cublas

4/6/0

714633.5

706840.5

0.9890951095911401

0.9675266011006094

1.0156027428769214

supported_no_material_regression

714060.5 | 702334.5 | 708082.5 | 732751.5 | 727409.5 | 712332.5 | 729704.5 | 698036.5

704442.5 | 700369.5 | 716407.0 | 725058.0 | 728021.5 | 706998.0 | 684640.5 | 702757.0

dense-256

cuda-native

4/6/0

702900.0

696544.0

0.9909574619433774

0.9604072866721649

1.0144506719259807

supported_no_material_regression

716347.0 | 671606.0 | 713990.0 | 736527.0 | 707760.5 | 698834.5 | 703445.0 | 693032.5

697516.0 | 690957.0 | 665219.0 | 685572.0 | 713111.5 | 702961.5 | 669654.0 | 715802.5

dense-256

torch-cuda-eager

4/6/0

1864800.5

1857808.0

0.9962502691306657

0.9546325204856435

1.0298531639155375

supported_no_material_regression

1942733.5 | 1751998.0 | 1858446.5 | 1995873.0 | 1871395.5 | 1846424.0 | 1850750.0 | 1812491.0

1861021.0 | 1931230.5 | 1739540.0 | 1864262.0 | 1902747.5 | 1832690.5 | 1820889.0 | 1875336.5

dense-64

cuda-cublas

5/7/0

777175.0

800526.0

1.0300459999356644

0.9865628668132439

1.0707614032686679

inconclusive

777315.5 | 713582.5 | 757393.0 | 844784.0 | 806248.0 | 796212.0 | 766137.5 | 788571.0

782663.0 | 823778.5 | 770991.5 | 799007.0 | 819414.0 | 807522.0 | 819068.0 | 782335.0

dense-64

cuda-native

5/7/0

733266.0

739794.5

1.0089033174864237

0.9765648044470041

1.04967256894527

supported_no_material_regression

749562.0 | 647113.0 | 718208.5 | 752703.0 | 730894.0 | 749570.0 | 726991.0 | 747028.5

725972.0 | 766522.5 | 711275.5 | 757741.5 | 757486.0 | 742318.0 | 729278.5 | 724308.5

dense-64

torch-cuda-eager

5/7/0

2208371.0

2213493.5

1.0023195830773

0.9686542219340932

1.0397571010006188

supported_no_material_regression

2229614.0 | 2027098.0 | 2230257.0 | 2321041.0 | 2210222.5 | 2261079.5 | 2210710.0 | 2099712.5

2221695.0 | 2305875.5 | 2135236.0 | 2147634.5 | 2257040.5 | 2222406.5 | 2187580.0 | 2220933.5

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused0-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

4/6/0

1054563.5

1048319.0

0.9940785927068403

0.9707446712281108

1.0208231570424078

supported_no_material_regression

1076131.0 | 1082285.0 | 1031772.5 | 1021843.5 | 1062003.0 | 1051984.5 | 1086262.0 | 1011030.0

1017265.5 | 1015415.0 | 1082767.0 | 1035672.0 | 1056396.0 | 1053759.5 | 1048887.0 | 1082256.0

dense-1024

cuda-native

4/6/0

1187114.5

1188779.0

1.00140213938925

0.9769240592533827

1.0229518472982275

supported_no_material_regression

1214025.0 | 1213315.5 | 1169647.0 | 1158313.5 | 1203587.5 | 1186588.0 | 1205960.0 | 1147680.5

1156334.5 | 1156521.5 | 1221031.5 | 1170675.5 | 1192217.0 | 1195409.0 | 1184087.0 | 1226236.5

dense-1024

torch-cuda-eager

4/6/0

1860650.0

1839287.0

0.9885185284712332

0.9362141772386586

1.0479522173029547

supported_no_material_regression

1952809.0 | 1901060.5 | 1782222.0 | 1757274.0 | 1909766.0 | 1876379.5 | 2070859.5 | 1763223.5

1774194.0 | 1744207.0 | 1840926.0 | 1767740.0 | 1866621.0 | 1875372.5 | 1889592.5 | 2037447.0

dense-2048

cuda-cublas

3/5/0

1388564.0

1379292.0

0.9933225980221293

0.9676462317670905

1.023208715491642

supported_no_material_regression

1390850.0 | 1539733.0 | 1373523.5 | 1362212.0 | 1389764.0 | 1384843.5 | 1406190.0 | 1356761.0

1444190.0 | 1368116.0 | 1419987.0 | 1350622.0 | 1372597.5 | 1388156.5 | 1327519.5 | 1390929.0

dense-2048

cuda-native

3/5/0

1575791.5

1559792.0

0.9898466897429006

0.955937891565377

1.010235859257088

supported_no_material_regression

1564497.5 | 1762318.5 | 1559240.5 | 1554643.5 | 1578745.0 | 1554497.5 | 1592335.5 | 1558941.5

1555085.0 | 1526089.5 | 1610675.5 | 1545395.5 | 1558299.0 | 1562089.5 | 1522849.5 | 1554436.0

dense-2048

torch-cuda-eager

3/5/0

1999834.5

1990629.0

0.9953968690909173

0.9377434318099894

1.0249273535477454

supported_no_material_regression

2007298.5 | 2630740.5 | 1927149.5 | 1902876.5 | 2006257.0 | 1999758.5 | 2178939.5 | 1943565.5

2019694.5 | 1985752.5 | 2003086.0 | 1925355.5 | 1961435.5 | 1967935.0 | 1961566.0 | 2059473.0

dense-256

cuda-cublas

4/6/0

862270.5

866835.5

1.0052941623307303

0.9774276406716034

1.0327623472273233

supported_no_material_regression

873757.0 | 874754.0 | 838767.0 | 841550.5 | 869854.5 | 867031.5 | 864360.0 | 897381.0

868881.0 | 826112.0 | 888493.0 | 837736.0 | 890648.5 | 878000.5 | 852740.0 | 883177.0

dense-256

cuda-native

4/6/0

884311.5

880071.5

0.9952053094412998

0.9697285354560797

1.0209219928984294

supported_no_material_regression

874394.0 | 900990.0 | 856865.5 | 865884.5 | 887396.5 | 888723.5 | 901114.0 | 897741.5

880011.5 | 849565.5 | 906766.5 | 841830.0 | 903639.5 | 871912.5 | 883729.0 | 909906.5

dense-256

torch-cuda-eager

4/6/0

1842131.5

1858584.0

1.008931229936625

0.9632080828052487

1.0481938369608184

supported_no_material_regression

1883951.5 | 1909118.5 | 1766551.0 | 1721827.5 | 1890490.0 | 1870275.0 | 1937019.0 | 1852211.0

1843948.0 | 1772333.0 | 1893171.0 | 1752177.0 | 1836351.5 | 1865552.5 | 1866545.0 | 2024005.0

dense-64

cuda-cublas

5/7/0

970125.5

970767.0

1.0006612546521043

0.9500959069948978

1.0552985073502457

inconclusive

999727.0 | 1027062.0 | 892637.0 | 845456.5 | 999121.5 | 978047.0 | 1006908.5 | 968582.0

969609.5 | 939091.0 | 1009738.5 | 868206.0 | 1003379.0 | 976157.0 | 959778.5 | 1025592.0

dense-64

cuda-native

5/7/0

954922.0

955443.0

1.0005455942998487

0.9499633353218847

1.0413770278266126

supported_no_material_regression

953598.0 | 987517.0 | 891278.0 | 833580.5 | 975053.5 | 975008.5 | 1013043.0 | 973842.5

944558.5 | 917674.0 | 999141.5 | 842173.0 | 999794.0 | 959032.0 | 944701.0 | 999841.5

dense-64

torch-cuda-eager

5/7/0

2217983.5

2222890.0

1.0022121444997223

0.9457132303012582

1.0655647156979726

inconclusive

2295969.0 | 2255137.0 | 2102385.5 | 2007024.5 | 2303581.0 | 2246831.5 | 2465110.5 | 2227331.5

2185644.0 | 2217690.5 | 2342504.0 | 1979733.0 | 2216999.0 | 2256737.5 | 2247512.5 | 2504912.5

cuda-release-cuda-csr-f64-v1-none-prepared-fused1-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

627/641/12

88252519.5

88003693.0

0.9971805167556718

0.9442259092441523

1.0470042601849727

supported_no_material_regression

82759628.5 | 93129892.5 | 85195134.0 | 85114543.0 | 90313729.5 | 87652846.0 | 93926613.5 | 92474332.0

85224832.0 | 84154545.5 | 89648855.5 | 85208703.5 | 86920000.5 | 89476433.5 | 92900849.0 | 94532696.0

poisson-128

torch-cuda-eager

627/641/12

168886530.0

167690333.5

0.9929171586389987

0.9597722813592648

1.041351338065685

supported_no_material_regression

157343677.5 | 178969666.5 | 169560247.5 | 160049599.0 | 175302328.5 | 167560462.0 | 172902006.5 | 171405442.0

161468578.5 | 163118303.5 | 167636133.0 | 165529656.5 | 168080225.5 | 173384615.0 | 173071161.0 | 178596253.5

poisson-32

cuda-native

118/122/2

14912465.5

15029629.0

1.0078567491069803

0.9781002693787123

1.0505163349858344

inconclusive

15112046.5 | 14780810.5 | 14178916.5 | 14059932.5 | 14496490.0 | 15242121.5 | 15240158.5 | 15463174.0

15095987.0 | 14334790.5 | 14340846.0 | 14977841.5 | 15974579.0 | 15147993.0 | 15100772.0 | 15395800.5

poisson-32

torch-cuda-eager

118/122/2

29611425.0

29524409.0

0.9970614045085638

0.9630260284233746

1.0500212509918632

inconclusive

32321190.5 | 28802098.5 | 27949888.5 | 27245200.0 | 29798949.5 | 30296825.0 | 30001435.0 | 30064564.0

29309752.0 | 28869939.5 | 27783021.0 | 28861794.0 | 31423736.0 | 31166160.5 | 29871728.0 | 30055404.5

poisson-64

cuda-native

201/207/4

25103916.5

25342735.5

1.0095132167922882

0.9655381061274889

1.0579429476363258

inconclusive

26321830.5 | 24627819.5 | 24126960.0 | 23821643.5 | 24327853.0 | 25320096.5 | 25801754.0 | 26252817.5

25645161.0 | 23825731.5 | 24383557.0 | 25019445.5 | 24568357.0 | 25024772.0 | 26597031.0 | 26431439.0

poisson-64

torch-cuda-eager

201/207/4

50187679.0

49992465.0

0.9961103202242128

0.9589225830857104

1.039084821406757

supported_no_material_regression

53075360.5 | 47604481.5 | 49588482.5 | 47336363.0 | 49690263.5 | 50970309.0 | 50912509.0 | 50600531.5

49861495.5 | 47228064.5 | 47312975.0 | 49507881.0 | 50142180.0 | 47505346.0 | 52991041.0 | 53923670.5

poisson-8

cuda-native

26/28/0

3296571.0

3366659.5

1.0212610315385289

0.9780787226784826

1.059455736776208

inconclusive

3218242.5 | 3638422.5 | 3307563.0 | 3233451.0 | 3193675.0 | 3354986.5 | 3367865.0 | 3400835.5

3511706.0 | 3455239.5 | 3233510.0 | 3733000.0 | 3327650.5 | 3404395.0 | 3256424.0 | 3288765.0

poisson-8

torch-cuda-eager

26/28/0

6619713.0

6720411.0

1.0152118377337507

0.9679154887259379

1.0605050509290261

inconclusive

6324652.5 | 7075379.5 | 6699255.0 | 6318075.5 | 6631950.5 | 6527247.0 | 6686202.0 | 6847306.0

6851362.5 | 6808792.0 | 6489216.0 | 7340509.0 | 6645111.5 | 6865132.0 | 6397621.5 | 6434798.0

cuda-release-cuda-csr-f64-v1-none-prepared-fused0-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

627/641/12

101922471.5

103626149.5

1.0167154306104125

0.9562474704809671

1.071151945752439

inconclusive

91952119.5 | 96489682.0 | 98914968.0 | 102930194.0 | 104461825.5 | 111200339.0 | 107710711.0 | 101529122.0

96441148.0 | 96463540.5 | 96186459.5 | 104486099.0 | 107791137.5 | 103626149.5 | 105978709.5 | 105065488.5

poisson-128

torch-cuda-eager

627/641/12

168326661.0

168460513.0

1.0007951919155575

0.9693044596921914

1.0361102600407177

supported_no_material_regression

155992307.0 | 163474002.5 | 164213184.5 | 169112850.0 | 171480537.5 | 177753976.0 | 172818633.0 | 165027598.0

156891145.5 | 165615574.0 | 168606564.0 | 170817927.0 | 174587304.0 | 166977039.5 | 169490946.0 | 169386635.0

poisson-32

cuda-native

118/122/2

15827036.5

15895275.5

1.004311546258202

0.946179955969812

1.0697002787367742

inconclusive

14476960.0 | 15169732.5 | 15992514.5 | 14719459.0 | 17856549.5 | 16099342.0 | 16696849.0 | 16049654.0

16726874.0 | 14920605.5 | 15768942.0 | 16497875.5 | 16173864.5 | 15700206.0 | 16556965.0 | 16613162.0

poisson-32

torch-cuda-eager

118/122/2

28407640.0

28993431.0

1.0206208963504184

0.9639064538987417

1.067847628254891

inconclusive

26737090.0 | 28303204.5 | 28742870.0 | 27689400.5 | 32532110.0 | 27488112.0 | 30675555.5 | 28817426.0

30474322.0 | 28009206.5 | 30441864.0 | 28540333.5 | 28649017.5 | 28873282.0 | 30286471.0 | 29461427.5

poisson-64

cuda-native

201/207/4

26475381.0

26785099.5

1.0116983585618655

0.9523696647406557

1.0588853891337446

inconclusive

25036731.0 | 25227546.5 | 25120836.0 | 26647847.5 | 29209390.0 | 26783160.0 | 28005952.5 | 27083799.0

26742833.0 | 25470546.5 | 25309095.5 | 25905552.5 | 26995258.5 | 27717569.0 | 27792903.5 | 27902610.5

poisson-64

torch-cuda-eager

201/207/4

48535106.0

48397024.5

0.9971550180605354

0.9539660402152235

1.0263748502835222

supported_no_material_regression

46281035.5 | 47470969.5 | 47004635.0 | 49885262.0 | 54698017.5 | 48402063.5 | 49879631.0 | 48455863.5

48495814.5 | 46268881.0 | 47610834.0 | 48033999.0 | 48480336.0 | 49410688.5 | 49474017.5 | 51880076.5

poisson-8

cuda-native

26/28/0

3546750.5

3641757.0

1.0267869138243584

0.946223861544252

1.0772393755202554

inconclusive

3296138.0 | 3416460.0 | 3564711.0 | 3480444.0 | 4306656.5 | 3480322.0 | 3819855.5 | 3620072.5

3960846.5 | 3413424.0 | 3508499.0 | 3342335.0 | 3673007.0 | 3707940.0 | 3711052.0 | 3725030.5

poisson-8

torch-cuda-eager

26/28/0

6499593.5

6594443.5

1.014593220329856

0.9250785415713507

1.0729584778770023

inconclusive

6108732.0 | 6356720.5 | 6471951.0 | 6430637.0 | 8107417.5 | 6355522.0 | 7046153.0 | 6508267.0

7609145.0 | 6424642.0 | 6825094.5 | 5962634.0 | 6618786.0 | 6576154.0 | 6640082.5 | 6812647.0

cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

627/641/12

58457528.5

58086342.5

0.9936503302564357

0.9712937176099452

1.0166333933048528

supported_no_material_regression

57675534.0 | 59218867.0 | 59011046.0 | 58818699.0 | 55379323.5 | 58231320.0 | 57393108.0 | 59100353.5

59117441.0 | 58478880.0 | 57682806.0 | 58304897.0 | 58133970.5 | 56330536.5 | 58979914.0 | 56383225.5

poisson-128

torch-cuda-eager

627/641/12

152304983.0

150843875.5

0.9904066993001798

0.9792698460091589

1.0092802502359373

supported_no_material_regression

149024498.0 | 153172697.5 | 153466612.0 | 153355873.5 | 150551352.5 | 149477440.5 | 152623172.5 | 152741912.5

150032540.0 | 150605120.0 | 148696135.0 | 154773675.5 | 155456807.5 | 147214828.5 | 154062897.5 | 150089600.0

poisson-32

cuda-native

118/122/2

9923449.5

9842746.5

0.991867444883959

0.9456731736754913

1.0273010032682657

supported_no_material_regression

10974812.0 | 9448366.0 | 9833169.5 | 9911130.5 | 9838754.5 | 9588152.0 | 9765461.5 | 10308296.5

9663481.5 | 10067256.5 | 9913787.5 | 9612133.0 | 9449510.5 | 10091891.5 | 10190517.0 | 9733374.0

poisson-32

torch-cuda-eager

118/122/2

29413374.5

28988230.0

0.985545878117453

0.9413758022456908

1.04357193650337

supported_no_material_regression

32965321.0 | 27941441.0 | 28896198.0 | 29871849.5 | 30298953.5 | 27626479.0 | 28565336.0 | 30143527.0

28685191.5 | 30086948.0 | 28982016.0 | 27796426.5 | 27917526.0 | 31161887.5 | 30913545.0 | 28095429.5

poisson-64

cuda-native

201/207/4

16839045.0

16832881.5

0.9996339756797372

0.9772044363386424

1.0255502467279694

supported_no_material_regression

16077983.5 | 17037483.0 | 16960069.0 | 16876985.0 | 17083310.5 | 16783525.0 | 16997448.0 | 17101054.0

16783791.5 | 18164964.0 | 16892513.5 | 17103375.5 | 16516619.5 | 16139690.0 | 17260833.0 | 16563057.0

poisson-64

torch-cuda-eager

201/207/4

50474658.5

50007560.0

0.9907458809255738

0.9567154593209315

1.0412120924892458

supported_no_material_regression

45671021.5 | 51041401.5 | 50361362.5 | 50831930.5 | 51611656.5 | 50096898.5 | 50084104.5 | 50698031.5

49939732.0 | 50561659.0 | 51314484.5 | 51239544.5 | 50311753.0 | 47611912.0 | 52925193.5 | 47309624.0

poisson-8

cuda-native

26/28/0

2331275.5

2261046.0

0.9698750748249189

0.932311662957203

1.0250886142655755

supported_no_material_regression

2591819.0 | 2196048.0 | 2212554.0 | 2396494.0 | 2336362.0 | 2211545.0 | 2332444.0 | 2493271.0

2376934.5 | 2194969.0 | 2303604.5 | 2272111.0 | 2194011.5 | 2262123.0 | 2439637.0 | 2194732.0

poisson-8

torch-cuda-eager

26/28/0

6733883.5

6521700.5

0.9684902478636584

0.9020711510947533

1.056991229716427

inconclusive

7390378.0 | 6378147.5 | 6736424.0 | 6827763.0 | 7065269.0 | 6146850.0 | 6498003.0 | 7069502.0

6888822.5 | 5971795.5 | 6949685.0 | 6732863.0 | 6431179.5 | 6522296.0 | 7353787.0 | 6165397.5

cuda-release-cuda-csr-f64-v1-none-prepared-fused0-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

627/641/12

61335016.0

61837011.5

1.008184484699572

0.9796322916438799

1.0300811721193066

supported_no_material_regression

61359182.5 | 60757150.5 | 60730792.5 | 62894766.5 | 60010262.5 | 65590557.0 | 61339917.0 | 59871283.0

60664787.0 | 61484270.0 | 61893703.5 | 62891634.5 | 61111979.0 | 63176463.0 | 61749848.0 | 61079007.5

poisson-128

torch-cuda-eager

627/641/12

148205203.0

148805844.0

1.004052765947765

0.9864457093592411

1.0200554909949509

supported_no_material_regression

149190662.0 | 150253234.0 | 146034545.0 | 151023857.5 | 143551312.0 | 152570360.5 | 150817858.5 | 144736708.0

147974290.5 | 149549166.5 | 149369757.0 | 149075935.5 | 147627462.5 | 149214882.5 | 149093625.5 | 148241501.5

poisson-32

cuda-native

118/122/2

11875793.5

11551221.0

0.9726694052064816

0.9319175123829295

1.0135899939260633

supported_no_material_regression

12157300.0 | 11725679.0 | 11197626.5 | 12688125.5 | 11923136.5 | 11748208.0 | 11274272.5 | 12001306.5

11550653.0 | 11738777.0 | 12099453.0 | 11158420.0 | 10999702.5 | 11932024.0 | 11539372.0 | 10995696.0

poisson-32

torch-cuda-eager

118/122/2

29712456.0

28928910.5

0.9736290564468989

0.9222057900375452

1.0452159657615485

supported_no_material_regression

31456377.5 | 30174474.5 | 27446711.5 | 31837277.0 | 30209596.5 | 28365724.5 | 26954637.5 | 30072966.0

29756353.5 | 30249022.0 | 29967509.0 | 28449216.0 | 26895385.0 | 30877225.0 | 28987712.5 | 26978281.5

poisson-64

cuda-native

201/207/4

19553419.5

19670275.0

1.0059762181238938

0.9715486578295909

1.0284562936534034

supported_no_material_regression

19900465.5 | 19365916.0 | 19255054.0 | 21742122.5 | 19482290.5 | 19644698.0 | 19292665.0 | 19149339.0

19299722.5 | 20106757.0 | 20219557.0 | 19908998.0 | 19370138.0 | 19584358.0 | 19663941.5 | 18984201.0

poisson-64

torch-cuda-eager

201/207/4

48629514.0

48292230.5

0.9930642222745636

0.9506343233417722

1.0287803167261036

supported_no_material_regression

51653470.0 | 48455201.5 | 47660914.5 | 53993122.0 | 47629155.5 | 48760145.0 | 47699538.5 | 46021075.0

47074036.5 | 51035771.0 | 51110111.0 | 47704941.0 | 47098722.5 | 49081789.0 | 48521113.5 | 46045774.0

poisson-8

cuda-native

26/28/0

2778153.5

2679337.5

0.9644310510560341

0.9150483128351375

1.018041722762678

supported_no_material_regression

3101924.5 | 2707355.0 | 2700750.0 | 3006234.5 | 2598969.0 | 2745464.5 | 2581898.5 | 2848068.0

2557861.0 | 2756254.0 | 2828852.5 | 2589398.0 | 2567612.0 | 2728673.0 | 2801736.0 | 2756287.0

poisson-8

torch-cuda-eager

26/28/0

6727190.5

6531968.5

0.970980158804779

0.9012038158451453

1.0556743690070314

inconclusive

7755797.5 | 6654025.5 | 6647808.0 | 7550128.0 | 6013441.0 | 6499389.0 | 6139102.5 | 6751780.5

6331579.5 | 7514237.0 | 6958088.0 | 6257924.5 | 6315537.0 | 6595852.0 | 6981188.5 | 6351302.5

cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

627/641/12

152289515.0

150585297.5

0.9888093576238653

0.9751462585411554

1.0094693554345784

supported_no_material_regression

145911116.0 | 154699874.5 | 150868831.0 | 153811760.0 | 151499382.0 | 151367263.0 | 153736985.5 | 152980873.0

144870928.0 | 150778360.5 | 152015361.0 | 149662251.5 | 156696124.0 | 148435192.5 | 151393981.0 | 152422115.0

poisson-128

torch-cuda-eager

627/641/12

254888870.5

253511556.5

0.994596413733961

0.9828503946781992

1.007365506145963

supported_no_material_regression

250316870.0 | 261306300.5 | 252433712.0 | 255242471.5 | 252476532.0 | 253157328.5 | 256496510.0 | 257397157.5

251758256.0 | 251757267.5 | 255240195.5 | 254314147.5 | 268739668.0 | 248027024.0 | 253224624.5 | 254077089.0

poisson-32

cuda-native

118/122/2

21028781.5

20859234.5

0.9919373835331353

0.9706947084893572

1.0196243797494942

supported_no_material_regression

21117593.5 | 20778735.0 | 21341997.0 | 21083754.5 | 20813014.0 | 21277461.5 | 21254914.0 | 20380540.5

20567428.5 | 20499674.0 | 20450662.0 | 22345182.0 | 20833044.5 | 21229623.0 | 21442867.5 | 20094248.0

poisson-32

torch-cuda-eager

118/122/2

40598667.5

39834524.5

0.9811781261047545

0.9554992719975688

1.0068443870218406

supported_no_material_regression

42632547.0 | 39777695.5 | 41929564.5 | 39876118.0 | 40355424.5 | 41283704.5 | 40656560.0 | 39509338.0

39541980.0 | 39319652.5 | 40094117.0 | 41550598.5 | 40367420.5 | 40984428.5 | 39799719.0 | 37910781.5

poisson-64

cuda-native

201/207/4

36422863.5

36638287.0

1.0059145130091158

0.9743160216020106

1.0406076803217208

supported_no_material_regression

35004701.5 | 36725440.0 | 36982216.5 | 36422599.5 | 35242482.5 | 38044525.5 | 37479968.0 | 35595445.5

36413021.5 | 34706976.0 | 37789753.0 | 38165785.5 | 36704433.0 | 36383529.5 | 36381208.0 | 35746759.5

poisson-64

torch-cuda-eager

201/207/4

69230827.5

70405082.0

1.0169614396130104

0.9723041675754388

1.0564335548882504

inconclusive

67567299.5 | 69880064.0 | 69888533.5 | 70176943.0 | 66892745.0 | 74240113.5 | 70609732.5 | 67072457.5

72801830.0 | 67694757.5 | 73267831.5 | 71171843.0 | 71054481.0 | 71686734.0 | 69200601.5 | 66896319.5

poisson-8

cuda-native

26/28/0

4722546.0

4708020.0

0.9969241167793813

0.9629643480996611

1.0323212087445814

supported_no_material_regression

4665912.5 | 4618543.5 | 5782831.0 | 4941995.0 | 4543911.5 | 4721561.0 | 4722044.0 | 4736600.0

4601870.5 | 4617973.5 | 4500042.0 | 4726303.5 | 4890270.0 | 5093817.5 | 4786821.5 | 4632390.0

poisson-8

torch-cuda-eager

26/28/0

9231814.0

9139751.5

0.9900276911991511

0.9460214821210606

1.0441172094643505

supported_no_material_regression

9377680.0 | 8990719.5 | 11471275.5 | 9809841.5 | 8491403.5 | 9152733.0 | 9103677.5 | 9241501.0

8907460.0 | 8847548.0 | 8922821.0 | 9007052.5 | 9656348.5 | 10325894.5 | 9177613.5 | 8780904.0

cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused0-host-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

627/641/12

155181341.0

152242990.5

0.9810650527887885

0.9382358921396751

1.0435936588822057

supported_no_material_regression

162827516.5 | 162714557.0 | 150048405.0 | 145485164.5 | 159911964.0 | 134896399.5 | 156123032.5 | 154210177.0

157084298.5 | 166365448.5 | 146146794.0 | 153150525.5 | 144653214.5 | 141205132.5 | 149710652.0 | 153962863.5

poisson-128

torch-cuda-eager

627/641/12

239810479.0

238078605.5

0.9927781575383118

0.9603624971286056

1.0312670014283847

supported_no_material_regression

246562640.5 | 245133076.5 | 231868756.0 | 236379312.0 | 252259687.0 | 216538309.5 | 242293193.5 | 239703043.5

239591149.5 | 251686294.5 | 235686255.5 | 236072858.5 | 222109181.0 | 235199558.0 | 237255431.5 | 248031777.5

poisson-32

cuda-native

118/122/2

23348908.5

22567316.0

0.9665255230239135

0.9236337918117229

0.999608294229844

supported_no_material_regression

24065095.5 | 22998191.0 | 23947891.0 | 21644766.0 | 28191140.0 | 24371236.5 | 22716244.5 | 22711230.0

22647886.5 | 22697538.0 | 22159510.5 | 22715783.0 | 22933160.5 | 24378465.5 | 21695339.5 | 21677859.5

poisson-32

torch-cuda-eager

118/122/2

40759886.0

39980415.5

0.9808765289480937

0.917300810405421

1.013102196074507

supported_no_material_regression

40588969.5 | 39651740.5 | 45013587.0 | 39737384.0 | 49502098.0 | 44024501.0 | 39565995.5 | 39334784.5

40080422.0 | 39632991.5 | 40629371.5 | 39407788.5 | 39902258.5 | 45824475.0 | 40135577.0 | 38665058.5

poisson-64

cuda-native

201/207/4

40646888.0

39420294.0

0.9698231756389321

0.9257568361211858

1.0393766345270936

supported_no_material_regression

41436814.0 | 39895641.0 | 40038430.5 | 41296893.0 | 42864205.0 | 42860880.5 | 37272609.0 | 38243465.0

40255536.0 | 42219821.0 | 37943079.5 | 38008701.5 | 40076925.0 | 44879427.0 | 38758686.0 | 38502113.0

poisson-64

torch-cuda-eager

201/207/4

71534768.0

70230671.0

0.9817697458667931

0.9254511063272919

1.0389829202547773

supported_no_material_regression

70201939.0 | 68879187.0 | 72604632.0 | 76862141.5 | 76282366.0 | 74139616.5 | 67975326.0 | 68504879.5

68340288.0 | 73424565.0 | 70127780.0 | 68467060.5 | 68372923.5 | 79015480.5 | 73085447.0 | 66818571.5

poisson-8

cuda-native

26/28/0

5228172.5

5131868.0

0.9815797011288361

0.9234986975626954

1.0348795209353514

supported_no_material_regression

5229110.5 | 5137779.5 | 5925562.5 | 4955936.5 | 5799889.5 | 5461099.0 | 4896943.5 | 4971596.0

5165822.0 | 5199510.5 | 5148973.0 | 4891161.5 | 5196186.5 | 5485921.5 | 5059818.0 | 4992828.0

poisson-8

torch-cuda-eager

26/28/0

9311298.0

9093117.5

0.9765681970440641

0.9045698832744609

1.0340063685452514

supported_no_material_regression

9051181.5 | 9033307.0 | 11324834.0 | 9085316.5 | 12029767.0 | 9701062.5 | 8915303.0 | 8618670.5

8867814.5 | 9086252.5 | 9071696.0 | 8952887.5 | 9140690.5 | 11338208.0 | 9363423.0 | 8955846.0

cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

627/641/12

65538704.0

66429028.0

1.0135847056115117

0.9660947839285017

1.0588335826121205

inconclusive

63645001.0 | 64911526.0 | 66311508.5 | 65009521.0 | 71447601.5 | 64829094.5 | 68309979.5 | 61978693.5

62551434.0 | 63891326.0 | 64541160.5 | 65288160.0 | 68835582.5 | 72873866.0 | 69307101.5 | 68223343.0

poisson-128

torch-cuda-eager

627/641/12

226229833.0

231465145.0

1.0231415632968266

0.9189722451008413

1.169766590173235

inconclusive

229948647.0 | 223270390.0 | 219036501.5 | 221760463.5 | 273249466.5 | 249372522.0 | 218710102.5 | 209657037.5

211884066.5 | 222246482.5 | 217682225.5 | 216101128.5 | 267943869.0 | 283235362.0 | 236710341.5 | 258591927.5

poisson-32

cuda-native

118/122/2

11897622.5

11591677.0

0.9742851565512354

0.8731720072173002

1.1110071296760506

inconclusive

10753340.0 | 12351071.5 | 11426770.5 | 11315869.0 | 15437421.0 | 15196805.5 | 10905072.5 | 12053274.5

10982937.5 | 11199864.0 | 11438527.0 | 11404696.0 | 14393127.5 | 17488201.0 | 11247542.0 | 11334352.5

poisson-32

torch-cuda-eager

118/122/2

42590682.5

41730613.0

0.9798061583070429

0.8354780204358967

1.2172599797779216

inconclusive

38683668.5 | 45030019.0 | 41066744.0 | 40020957.5 | 53642980.5 | 53051011.5 | 38506602.5 | 43914924.5

39818456.5 | 40482827.5 | 41987263.0 | 41690939.0 | 54842190.0 | 63712835.0 | 40028695.0 | 39812097.5

poisson-64

cuda-native

201/207/4

20237425.5

19936291.0

0.9851199205155814

0.8013855888305972

1.2068431257900687

inconclusive

18564985.5 | 20491538.0 | 19691051.5 | 19255775.5 | 26031638.0 | 27445983.5 | 18465032.5 | 19858412.0

19006737.0 | 19501173.0 | 18807400.0 | 19753251.0 | 25187573.5 | 32712586.0 | 22408652.5 | 19205218.5

poisson-64

torch-cuda-eager

201/207/4

70880342.5

71224196.0

1.0048511828226563

0.7904242694854567

1.2491851407315366

inconclusive

65362903.0 | 73163597.0 | 71189337.5 | 68011256.0 | 96132578.0 | 107895962.0 | 64385987.5 | 69934023.0

67831083.5 | 71039090.0 | 67296436.5 | 67512358.5 | 89847301.0 | 122569456.0 | 75602996.0 | 69244310.5

poisson-8

cuda-native

26/28/0

2709538.0

2682016.0

0.989842548803523

0.921865172494711

1.1138806162655062

inconclusive

2611624.5 | 3171710.5 | 2552970.5 | 2687836.5 | 2896984.5 | 3001799.5 | 2569898.5 | 2742577.5

2638654.0 | 2554428.5 | 2637626.0 | 2603807.5 | 3205911.0 | 4073819.0 | 2568139.5 | 2751041.5

poisson-8

torch-cuda-eager

26/28/0

9704052.0

9808201.0

1.0107325269897565

0.9141651509482278

1.1778712359759917

inconclusive

8514236.0 | 10301195.5 | 8920168.0 | 9555158.0 | 12488333.5 | 11540148.5 | 8968512.0 | 9653027.5

9539604.5 | 8945747.0 | 9652033.0 | 9521569.0 | 11694834.5 | 15424995.0 | 9083968.5 | 10862817.0

cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused0-device-standard

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

627/641/12

96601288.0

96111826.0

0.9949331731477534

0.9197862722938879

1.0608997027695015

inconclusive

88711169.0 | 101535748.0 | 99020527.0 | 94285898.5 | 95162073.0 | 93921917.5 | 98023160.5 | 102325151.5

88660852.5 | 85884236.0 | 96122212.0 | 95185990.5 | 98203041.5 | 96556905.0 | 104457040.0 | 100868911.5

poisson-128

torch-cuda-eager

627/641/12

266980333.5

266204878.5

0.9970954602167355

0.8360674725053606

1.078697523722567

inconclusive

215392077.0 | 274630777.0 | 279371902.0 | 254406549.0 | 263389637.5 | 253039467.0 | 294539800.0 | 295509635.0

214209646.5 | 201423420.0 | 285659706.0 | 256431103.0 | 269146926.0 | 265768362.0 | 278291739.5 | 281485217.5

poisson-32

cuda-native

118/122/2

19490238.0

19491376.5

1.0000584138582607

0.8608531820562849

1.1795310504627374

inconclusive

15963720.5 | 16351977.5 | 19656634.5 | 19385168.5 | 20105189.0 | 21013361.5 | 22362330.0 | 20588458.5

16237407.5 | 16299829.5 | 20781465.0 | 19346974.5 | 20630736.5 | 20401387.5 | 20300005.5 | 21563539.5

poisson-32

torch-cuda-eager

118/122/2

50436204.5

52511987.0

1.0411565961510842

0.8352729517355462

1.213919803706613

inconclusive

40094819.0 | 40837659.5 | 53736058.5 | 48031106.0 | 51955845.0 | 54292787.5 | 56760806.5 | 54766215.0

40350327.0 | 40876715.5 | 54674301.0 | 51855621.0 | 55599698.5 | 56052061.0 | 51384718.5 | 56283589.5

poisson-64

cuda-native

201/207/4

33632699.0

33800083.0

1.0049768232992542

0.8743104961633573

1.110327408964205

inconclusive

28376787.5 | 30365215.5 | 29896775.0 | 33930176.0 | 34367016.0 | 34962214.0 | 36619829.5 | 34404517.5

27300320.0 | 27813088.0 | 33901653.0 | 34331020.0 | 34190773.5 | 35805462.5 | 34196190.5 | 35794636.5

poisson-64

torch-cuda-eager

201/207/4

88159133.5

85667038.5

0.9717318569153133

0.8292452810817371

1.078899874936456

inconclusive

68220276.0 | 73677927.0 | 89015806.0 | 82651065.0 | 89262395.5 | 93382125.5 | 93783675.5 | 91581174.0

67470865.0 | 67623758.0 | 87032194.5 | 84636062.5 | 85065091.5 | 92992297.0 | 90634309.0 | 90304029.0

poisson-8

cuda-native

26/28/0

3913615.0

4002599.0

1.0227370346853228

0.8931266629431445

1.134396271453298

inconclusive

3528833.5 | 3728952.5 | 4363666.0 | 4347721.0 | 3789778.5 | 4982409.5 | 4876061.0 | 3609169.5

3630532.0 | 3804925.5 | 4125001.0 | 3944276.5 | 4967989.0 | 4126348.5 | 4460090.0 | 4342648.0

poisson-8

torch-cuda-eager

26/28/0

11606852.0

10883876.5

0.9377113191414864

0.8132769343538063

1.2051468890877253

inconclusive

8533787.5 | 8989983.5 | 11772710.5 | 11460639.5 | 12555724.5 | 12727214.0 | 13200844.5 | 12425543.0

8754384.5 | 9455770.5 | 11370849.5 | 12153374.5 | 10958017.0 | 14023432.5 | 12052912.5 | 11290743.5

cuda-release-cuda-dense-f64-v1-none-prepared-fused1-host-explicit

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

69/139/0

23745382.0

24068859.0

1.0136227330434187

0.980185510542196

1.0550484895798298

inconclusive

23213473.0 | 23973881.5 | 24813489.0 | 22454794.5 | 24230765.5 | 25217428.5 | 23268297.0 | 23013649.5

22975912.5 | 25262751.0 | 23709165.0 | 24229477.5 | 23400593.5 | 25960698.0 | 24001823.5 | 23571870.0

dense-1024

cuda-native

69/139/0

26143704.0

26497513.0

1.0135332392074206

0.9844551353645856

1.0514612673599188

inconclusive

25970999.5 | 26281301.5 | 26788444.5 | 25374554.0 | 26079396.0 | 27076853.0 | 26419987.5 | 25706116.0

24868016.0 | 27906765.5 | 26466082.0 | 26405178.5 | 26113297.0 | 28194835.0 | 26383777.5 | 26255955.5

dense-1024

torch-cuda-eager

69/139/0

39727369.5

40023736.5

1.0074600207295377

0.960671019668006

1.0592723265514852

inconclusive

39842215.0 | 42100365.5 | 41842733.5 | 38883490.5 | 39622081.5 | 38353242.0 | 39553900.5 | 37258568.0

38385113.0 | 43084297.0 | 39162410.5 | 39322245.5 | 37989257.0 | 42671592.5 | 39888413.5 | 41352759.5

dense-2048

cuda-cublas

68/137/0

37992027.0

37983515.5

0.9997759661520561

0.9799866830203003

1.0204795195485101

supported_no_material_regression

38167848.5 | 40147705.5 | 37681854.0 | 37527313.0 | 38027183.0 | 37876265.0 | 37977895.5 | 37082646.5

38976887.0 | 38891610.5 | 37198819.5 | 38461455.5 | 37227461.0 | 37991557.5 | 37781041.5 | 38136089.0

dense-2048

cuda-native

68/137/0

41034725.0

41190718.0

1.0038014876424783

0.9819583690158934

1.0275101496725414

supported_no_material_regression

41318279.5 | 42799454.5 | 40869966.5 | 40815652.5 | 40436274.5 | 40728310.5 | 41239714.0 | 39878777.5

43107178.5 | 42153387.0 | 40227976.0 | 41433719.0 | 40272492.5 | 42264276.0 | 40544849.0 | 41435338.0

dense-2048

torch-cuda-eager

68/137/0

52433318.5

52902401.0

1.0089462676294272

0.9716477413216517

1.0431155206438176

supported_no_material_regression

52415846.5 | 57025146.5 | 51455885.5 | 50860242.5 | 53019600.5 | 52833133.5 | 53137319.5 | 51250998.5

54542022.5 | 54633061.0 | 51479721.0 | 53694267.5 | 52447944.0 | 52337111.5 | 51502970.0 | 55433677.5

dense-256

cuda-cublas

65/131/0

20305370.5

19644603.5

0.9674585105452767

0.9399660217025115

0.9938504727017932

supported_no_material_regression

20183906.0 | 20269699.5 | 20912930.0 | 20156617.0 | 20198077.0 | 19989622.0 | 20717141.5 | 20467345.5

18927817.0 | 20765810.5 | 19619422.0 | 19963333.0 | 19490941.0 | 20310572.5 | 19847764.0 | 18925954.0

dense-256

cuda-native

65/131/0

19760942.0

19953455.5

1.0097421216053364

0.9656012515241603

1.0442784848888262

supported_no_material_regression

19811107.0 | 19330098.0 | 20860831.5 | 19235277.0 | 19550672.0 | 20194644.0 | 20220995.5 | 20391249.5

19293265.5 | 20961545.0 | 19031164.0 | 20022598.0 | 18961964.0 | 20725075.0 | 20760683.5 | 20022453.5

dense-256

torch-cuda-eager

65/131/0

38974594.0

39335164.0

1.0092514113168183

0.9689629117871718

1.058206707369132

inconclusive

40002472.5 | 37926447.5 | 42554220.5 | 37858006.5 | 38438663.5 | 38120134.5 | 38943188.0 | 39327192.5

39063558.5 | 41481230.0 | 37936702.0 | 40539234.5 | 37621024.5 | 39701196.0 | 43249551.5 | 37700891.0

dense-64

cuda-cublas

49/99/0

15193023.5

15334194.0

1.009291797646466

0.9643295413542061

1.075438029373457

inconclusive

15555327.0 | 14848233.0 | 15438100.0 | 15233969.5 | 14359398.0 | 16266064.0 | 14949254.5 | 16070190.0

15387089.0 | 18170859.0 | 14905877.5 | 14686946.5 | 14468687.0 | 15539715.0 | 16117879.0 | 15716100.0

dense-64

cuda-native

50/101/0

14841173.0

14901315.0

1.004052375105391

0.9657684079504731

1.0555717008347425

inconclusive

15198705.0 | 14536534.0 | 14715667.0 | 14943155.5 | 13761794.0 | 15421130.0 | 15058509.0 | 14906280.0

15162349.0 | 18857439.5 | 14590645.0 | 14454077.0 | 14237637.0 | 15274618.0 | 15001069.5 | 14668789.0

dense-64

torch-cuda-eager

49/99/0

29586810.0

30148329.0

1.018978693546212

0.9818699475819083

1.0721198896415374

inconclusive

30834634.5 | 28721433.0 | 29511644.5 | 29681043.5 | 27997250.0 | 29417833.5 | 29790833.0 | 30436457.5

30417589.0 | 34984672.0 | 28710883.5 | 29343541.5 | 28005020.5 | 29673383.0 | 31290460.5 | 30402391.5

cuda-release-cuda-dense-f64-v1-none-prepared-fused0-host-explicit

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

69/139/0

24034958.0

24292971.0

1.0107349053824017

0.9729948587756706

1.0503212180545098

inconclusive

23707862.5 | 23529232.0 | 23202854.5 | 25026042.0 | 24858439.5 | 23941995.5 | 24081502.0 | 24346621.5

25279392.0 | 25393079.5 | 25056893.0 | 23197990.5 | 26005265.0 | 23495249.5 | 22825964.5 | 24061281.5

dense-1024

cuda-native

69/139/0

26139390.0

26423084.0

1.010853122433232

0.9631369700748543

1.0448295146965114

supported_no_material_regression

25670929.0 | 26053139.5 | 26753693.5 | 27546927.0 | 26845151.5 | 25707703.0 | 25382315.5 | 26846491.5

27566259.5 | 26795501.0 | 27330861.5 | 25095783.5 | 27179815.5 | 26274817.0 | 24659356.5 | 25792233.5

dense-1024

torch-cuda-eager

69/139/0

39140085.5

39583379.0

1.0113258183863703

0.9671885618316841

1.0754079717523946

inconclusive

37963490.5 | 38195316.5 | 37768855.0 | 41710195.0 | 40190970.0 | 39969682.5 | 37771541.0 | 38750408.5

41806474.0 | 40320841.0 | 41973071.5 | 37075483.5 | 43181968.5 | 38662978.5 | 37292275.0 | 37366235.0

dense-2048

cuda-cublas

68/137/0

37875523.5

38551140.0

1.0178378128555767

0.9875540076816562

1.0581485282687768

inconclusive

38191493.5 | 38182731.5 | 35398607.0 | 37646985.5 | 38887791.5 | 36429830.5 | 37534517.0 | 38591917.5

39450725.0 | 39523042.0 | 38055772.5 | 41210953.5 | 38365243.0 | 37843197.0 | 36782461.0 | 37634307.0

dense-2048

cuda-native

68/137/0

40236555.0

40832329.0

1.0148067845271544

0.9880732114832128

1.0510109837639543

inconclusive

40474790.5 | 40398801.0 | 37628083.5 | 39193580.5 | 41852957.0 | 39392396.0 | 39549491.0 | 41083535.5

41079229.5 | 41840866.5 | 40270450.0 | 44059621.5 | 41358353.5 | 39764500.0 | 39502685.5 | 40017077.0

dense-2048

torch-cuda-eager

68/137/0

51168864.0

52791375.5

1.0317089607461287

0.9918543769428666

1.0789139007397328

inconclusive

51799610.0 | 51684655.5 | 47208752.0 | 49304871.5 | 54527147.5 | 50432993.5 | 50750967.0 | 54516342.5

52321123.5 | 55112539.0 | 51531854.5 | 57357339.0 | 54764023.5 | 52317400.0 | 51230724.5 | 51475630.5

dense-256

cuda-cublas

65/131/0

20747985.5

20119285.0

0.9696982388964943

0.9186932231135618

1.0140181083512267

supported_no_material_regression

20478606.0 | 20654654.0 | 23144930.5 | 18940456.0 | 22655307.5 | 20976857.0 | 20400906.0 | 18872839.0

19951280.5 | 19919320.0 | 20523108.0 | 20750219.5 | 19334455.5 | 20741874.0 | 19183708.5 | 20619541.5

dense-256

cuda-native

65/131/0

21643321.0

20917510.0

0.9664648969536607

0.9321075185336511

1.0001096322660439

supported_no_material_regression

20983540.5 | 21218310.0 | 23764875.0 | 21181308.0 | 22811913.5 | 21833598.0 | 21735103.0 | 21438434.0

20401523.5 | 20814088.5 | 21330425.0 | 21911303.5 | 21807834.0 | 21413083.0 | 19961805.5 | 20940921.0

dense-256

torch-cuda-eager

65/131/0

39650792.0

38123863.5

0.9614905926721464

0.917732788033863

1.017494048149651

supported_no_material_regression

39260762.0 | 38113292.5 | 44909364.0 | 37892403.5 | 41365082.0 | 40935993.0 | 40174564.5 | 37649660.5

38029008.5 | 38514254.5 | 39454675.5 | 41711032.5 | 40789749.0 | 37842980.5 | 36912902.5 | 36835107.0

dense-64

cuda-cublas

49/99/0

15657598.5

15142842.0

0.9671241729694372

0.9102702155478287

1.0410227479090473

supported_no_material_regression

14622658.5 | 15393559.0 | 17364266.0 | 16690962.0 | 15605160.5 | 15174615.0 | 15292622.0 | 13996935.0

15836397.0 | 14899620.0 | 14590032.5 | 15272181.5 | 14371821.0 | 15260665.5 | 14676851.5 | 16039191.5

dense-64

cuda-native

50/101/0

15637874.5

15159489.5

0.9694085663623915

0.9187738628580986

1.015251400241204

supported_no_material_regression

15280499.0 | 15679766.0 | 17455129.0 | 17229691.0 | 15852021.0 | 15082355.5 | 15147069.0 | 14191486.0

15528400.0 | 15721236.0 | 15060374.0 | 15645924.5 | 14096416.0 | 15016089.0 | 14323750.5 | 15995102.5

dense-64

torch-cuda-eager

49/99/0

30405149.5

29293634.5

0.9634431989883819

0.8997422648148798

1.0374260565513265

supported_no_material_regression

28554573.0 | 29665745.5 | 33278791.0 | 34360905.5 | 31108897.0 | 30448145.5 | 30072768.5 | 28749711.5

32160089.0 | 28543046.5 | 28437682.0 | 29562148.0 | 28528352.0 | 28273860.0 | 28256631.0 | 33555084.5

cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-explicit

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

69/139/0

17385447.0

17085809.0

0.982765010298556

0.9285052443595241

1.073648209192014

inconclusive

16534882.5 | 15258686.0 | 18802723.5 | 17090261.0 | 17490242.5 | 17370879.5 | 17455311.5 | 17841756.0

14724177.5 | 16548797.0 | 17610294.5 | 16159533.5 | 17151685.5 | 18246146.0 | 18821957.5 | 18460731.5

dense-1024

cuda-native

69/139/0

19092880.0

19110933.5

1.0009455619057994

0.9475127385998707

1.0591899003456064

inconclusive

19127337.5 | 16722942.0 | 18856337.5 | 20413003.0 | 19115117.5 | 19174248.0 | 19604231.5 | 19276888.0

16649458.5 | 18873100.0 | 18511043.5 | 17664331.0 | 19225052.0 | 21157494.5 | 19714326.0 | 19527243.0

dense-1024

torch-cuda-eager

69/139/0

38733167.5

38988091.5

1.0065815428082405

0.9443230569044749

1.0687759860371913

inconclusive

38129345.5 | 34313432.0 | 42945059.0 | 42435120.5 | 39994649.5 | 37192126.0 | 38887113.0 | 39958205.0

35513853.0 | 37934728.5 | 39604409.5 | 37333780.0 | 39017384.0 | 44892598.5 | 40141377.0 | 39644325.0

dense-2048

cuda-cublas

68/137/0

31198436.5

31673565.5

1.015229256760992

0.9629654140761185

1.067352861913659

inconclusive

29490268.5 | 30785462.0 | 28093441.0 | 29843830.5 | 31969789.0 | 31097094.0 | 32381790.0 | 32888998.0

32461895.0 | 29664541.0 | 31737254.5 | 26587114.5 | 31770790.5 | 31027220.5 | 32877569.5 | 33617633.5

dense-2048

cuda-native

68/137/0

34234144.5

34291120.5

1.001664303309814

0.954305643407778

1.0578658495590445

inconclusive

31698406.5 | 33880132.0 | 30931164.5 | 34561347.0 | 34962357.0 | 33634480.0 | 35668975.0 | 35125392.5

34949322.5 | 31974317.0 | 35279716.5 | 29589169.0 | 34555994.0 | 33658702.5 | 35595717.5 | 35851349.5

dense-2048

torch-cuda-eager

68/137/0

51941751.5

52916677.0

1.0187695923191962

0.9545026505579276

1.0929052665408725

inconclusive

47710917.5 | 52899523.0 | 48446926.0 | 50629391.0 | 53769531.5 | 51549597.0 | 54662763.0 | 54396003.5

55746850.5 | 47168211.5 | 54898794.5 | 44798182.0 | 53464238.0 | 51382916.0 | 54727104.0 | 58211816.0

dense-256

cuda-cublas

65/131/0

14310427.5

14529471.5

1.0153066007287344

0.9221217569174929

1.144376608861759

inconclusive

15147836.5 | 13385341.5 | 14993138.0 | 11278359.0 | 15395695.0 | 13050413.5 | 14348722.5 | 15391743.0

14720033.0 | 14452871.0 | 14159945.5 | 14828692.5 | 13823322.0 | 14753334.5 | 15627473.5 | 13315033.5

dense-256

cuda-native

65/131/0

13157183.0

13569319.0

1.0313240303794513

0.9288004776717516

1.1404456065154753

inconclusive

13585813.5 | 11710074.0 | 14140473.0 | 10733365.5 | 14074295.0 | 13044214.5 | 13878297.0 | 14320395.5

11998842.5 | 13464474.0 | 12305449.0 | 13523911.0 | 14305567.5 | 13872712.0 | 15075223.0 | 14259422.5

dense-256

torch-cuda-eager

65/131/0

39936834.5

39916490.0

0.9994905830606079

0.936131182387547

1.0943810415915796

inconclusive

42240282.0 | 35608190.0 | 44169639.0 | 33793703.5 | 44640886.5 | 37820893.5 | 40274060.0 | 41771570.0

37227209.0 | 39615612.5 | 38922255.0 | 40657864.0 | 39944756.0 | 41069417.5 | 42186815.0 | 39923290.0

dense-64

cuda-cublas

49/99/0

9946514.5

10130165.5

1.0184638548508627

0.9115272499019522

1.1550405717247187

inconclusive

9053537.0 | 9948678.0 | 11960989.0 | 8273758.0 | 10139138.5 | 9368345.5 | 10604173.5 | 10336259.5

11471025.5 | 8540672.5 | 9308801.0 | 9611007.5 | 10366871.5 | 11650574.0 | 10216162.5 | 10013970.5

dense-64

cuda-native

50/101/0

9093135.5

9233959.0

1.0154867922071544

0.9106961657674344

1.1540327625887141

inconclusive

8108802.0 | 9180044.5 | 10868287.5 | 7931381.0 | 9199678.0 | 8486371.5 | 8914122.0 | 9604733.0

11200661.5 | 7985685.5 | 8281544.0 | 9232663.5 | 9319839.0 | 10649426.0 | 8962490.5 | 9454637.5

dense-64

torch-cuda-eager

49/99/0

29553657.5

30070777.0

1.0174976481337377

0.9434235165136929

1.1252471036088882

inconclusive

28180337.5 | 28339209.0 | 35057497.5 | 25816955.0 | 30540135.0 | 28483777.0 | 29993412.0 | 32448555.5

29558866.0 | 28819529.5 | 27813186.0 | 27635185.5 | 32546359.0 | 34499619.5 | 30983491.0 | 30254333.5

cuda-release-cuda-dense-f64-v1-none-prepared-fused0-device-explicit

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

dense-1024

cuda-cublas

69/139/0

16484033.0

17222781.5

1.0448160046755548

0.8983456465187366

1.1985619066971236

inconclusive

13958059.0 | 14173364.5 | 14259971.0 | 24974615.5 | 17933196.5 | 17828899.0 | 17130879.5 | 18362263.5

17321529.5 | 17671286.5 | 15938498.0 | 16779192.0 | 16144019.0 | 18294066.5 | 19247543.5 | 17605449.0

dense-1024

cuda-native

69/139/0

18355893.5

19498252.0

1.0622338814506633

0.9361346540871844

1.1870103821656202

inconclusive

15869989.0 | 16048203.5 | 15900782.5 | 25831108.5 | 19868605.0 | 19374351.5 | 18883702.0 | 20179218.5

19069813.5 | 19609415.0 | 18025962.5 | 17969267.0 | 19721206.0 | 20905645.0 | 20986513.0 | 19630581.5

dense-1024

torch-cuda-eager

69/139/0

39158705.5

39671126.0

1.0130857364526515

0.8988572140137421

1.2481438332088608

inconclusive

31479122.0 | 31372048.5 | 30659677.5 | 56506296.5 | 43211747.5 | 39999518.5 | 39005591.5 | 42846820.0

38920009.0 | 42346430.5 | 36725500.5 | 32218336.5 | 38357893.0 | 43499544.5 | 44250803.0 | 37358940.5

dense-2048

cuda-cublas

68/137/0

32221730.5

31862306.0

0.9888452763267944

0.9455528658731259

1.0435087686881672

supported_no_material_regression

33491147.0 | 26304523.5 | 31522511.5 | 39387322.5 | 32196649.0 | 31459484.5 | 32641515.0 | 32119384.0

33501063.0 | 29960772.0 | 31238101.0 | 37955605.0 | 31606977.5 | 31440227.5 | 31912334.5 | 31644497.5

dense-2048

cuda-native

68/137/0

34511492.0

33949116.5

0.9837046888613219

0.9415776272861367

1.0385110126022057

supported_no_material_regression

34667238.0 | 28449054.0 | 34892523.0 | 43506225.0 | 34590233.0 | 34772242.0 | 33867809.0 | 33877537.5

36115176.5 | 32394855.0 | 31973581.5 | 40309759.0 | 33726700.0 | 33917614.5 | 33682735.5 | 34225731.5

dense-2048

torch-cuda-eager

68/137/0

53464826.5

52285272.0

0.9779377475394968

0.9189188669904885

1.058777910708251

inconclusive

55698743.0 | 42284322.0 | 54857875.5 | 64581349.0 | 54702612.0 | 52881009.0 | 52686060.0 | 51252742.0

54380380.0 | 49851905.5 | 49996943.0 | 62603569.5 | 51716106.5 | 52565147.5 | 51688475.5 | 51496502.5

dense-256

cuda-cublas

65/131/0

13828169.0

13767354.5

0.9956021292479141

0.8681298284287916

1.1517137887383053

inconclusive

11921759.5 | 12293386.5 | 12054931.0 | 16955450.5 | 15435179.0 | 14460775.5 | 14316066.0 | 15234742.0

12425931.5 | 16529724.0 | 12276956.0 | 14569225.0 | 15202379.5 | 14318800.5 | 13274630.5 | 15936391.5

dense-256

cuda-native

65/131/0

14375304.0

14050603.5

0.9774126168044863

0.8667183554338596

1.1419444946329171

inconclusive

11877544.0 | 13354688.0 | 12335497.5 | 16187905.0 | 15449175.5 | 15537628.5 | 13997451.0 | 15578419.5

12568543.5 | 15401428.0 | 12009773.5 | 14507048.0 | 13841811.5 | 14224625.5 | 15946303.0 | 14754396.5

dense-256

torch-cuda-eager

65/131/0

38533384.5

38360335.0

0.9955091019840212

0.8955854027553791

1.1133473425967195

inconclusive

32698603.0 | 35728490.0 | 33024116.5 | 47802693.5 | 41661870.0 | 39062868.5 | 37928580.0 | 42590038.0

32728941.5 | 45384233.0 | 33916149.0 | 41683235.0 | 37289386.0 | 38772338.5 | 41226888.5 | 37153512.0

dense-64

cuda-cublas

49/99/0

9982164.5

9603738.5

0.9620897852364585

0.8475684324761215

1.1064319979287185

inconclusive

8820086.5 | 8986226.5 | 8591711.0 | 12961848.0 | 10599807.5 | 10704054.0 | 9908685.0 | 10605430.0

8735108.5 | 10885711.5 | 8370055.5 | 10690727.0 | 9074601.5 | 9023334.5 | 9943728.5 | 10540477.5

dense-64

cuda-native

50/101/0

9877047.0

9832444.0

0.995484176596507

0.8503489133575478

1.147784074713243

inconclusive

8867146.5 | 8555614.5 | 9115143.0 | 13036405.5 | 10610780.5 | 11009415.5 | 9534389.0 | 10803775.0

9182530.0 | 10473829.5 | 8336602.5 | 11136842.5 | 8973685.0 | 8739115.5 | 10890255.5 | 10966659.0

dense-64

torch-cuda-eager

49/99/0

29064227.5

29363416.0

1.0102940461775562

0.8939355705685819

1.1055801231412297

inconclusive

25719808.5 | 26830276.0 | 23971215.0 | 33418965.0 | 31952977.5 | 29462881.0 | 28563404.5 | 30683629.0

25582714.0 | 32298570.5 | 23294413.0 | 31558013.0 | 28085670.5 | 28917467.5 | 31842455.5 | 30777169.5

cuda-release-cuda-csr-f64-v1-none-prepared-fused1-host-explicit

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

325/651/0

102356476.0

100392635.0

0.9808137103117931

0.9489827802552068

1.0114038199236857

supported_no_material_regression

102715245.5 | 99891125.0 | 100562418.5 | 105037977.5 | 107175460.5 | 100440764.5 | 102942084.0 | 97206986.5

102631076.5 | 102325180.0 | 94701320.0 | 102826109.5 | 97245702.0 | 97425947.0 | 101167030.0 | 100337689.0

poisson-128

torch-cuda-eager

325/651/0

192520984.5

186337668.0

0.9678823764793286

0.9242203573550796

1.0060990247973878

supported_no_material_regression

190201823.5 | 186294471.0 | 203261528.5 | 191260995.0 | 210263185.5 | 178210263.5 | 197210964.0 | 186375950.0

186800307.5 | 194239579.0 | 174726173.5 | 196935129.5 | 182262675.5 | 181078651.0 | 195678825.5 | 184975633.5

poisson-32

cuda-native

92/185/0

28041007.0

27964261.5

0.9972630975770592

0.9253760287071209

1.041073644534012

supported_no_material_regression

33023271.5 | 27397219.0 | 25314902.5 | 31492946.5 | 27810741.0 | 27781736.0 | 26790638.5 | 27794298.0

30723487.5 | 27425753.5 | 27081500.5 | 28274947.0 | 27890592.0 | 28321788.5 | 26720011.5 | 27927137.5

poisson-32

torch-cuda-eager

92/185/0

53868694.5

52718622.5

0.9786504571778698

0.8945649324371054

1.0481502639296127

supported_no_material_regression

63019405.0 | 51608570.0 | 49446662.0 | 59421377.0 | 53757243.5 | 52242801.0 | 51097615.0 | 51878722.0

59419336.0 | 50967934.5 | 49469414.5 | 53375541.5 | 51976315.5 | 56354611.0 | 51203573.0 | 53950452.0

poisson-64

cuda-native

174/349/0

54449585.0

54616439.0

1.0030643759727462

0.9224632046243066

1.0700074142030416

inconclusive

58031142.5 | 54792798.5 | 56193494.0 | 59045806.0 | 52397521.0 | 52624930.5 | 51490763.5 | 52335014.0

58686669.0 | 55159396.0 | 49010281.5 | 61622770.5 | 54807067.0 | 51411441.0 | 52090779.0 | 50459303.0

poisson-64

torch-cuda-eager

174/349/0

101455199.5

102875697.0

1.0140012291829361

0.9283875589597488

1.0927072861082079

inconclusive

109817440.5 | 98063842.0 | 111522975.0 | 110875638.0 | 96414555.0 | 102674385.0 | 96416755.0 | 97285289.0

109750797.0 | 103157858.0 | 90808326.0 | 122099682.0 | 103958393.0 | 93242133.0 | 100261961.0 | 99674014.0

poisson-8

cuda-native

26/53/0

7587425.5

7869466.5

1.0371721607019404

0.9183295711617001

1.1904913494810376

inconclusive

10204117.5 | 7013104.0 | 6483415.0 | 8013535.0 | 7449162.0 | 7389995.5 | 7969709.0 | 7985205.0

8725088.0 | 7878450.0 | 9394615.5 | 7730143.5 | 7391514.0 | 7867316.0 | 7645596.0 | 7380861.0

poisson-8

torch-cuda-eager

26/53/0

15424450.0

15649662.5

1.0146010068430318

0.891691785447355

1.1576741775166697

inconclusive

21161591.0 | 13529364.0 | 12405058.0 | 17103943.0 | 14485215.5 | 14467835.0 | 14976668.0 | 16917076.0

14883367.5 | 14520485.0 | 18358011.5 | 17071504.5 | 15337486.0 | 15557382.0 | 15300728.5 | 16143382.5

cuda-release-cuda-csr-f64-v1-none-prepared-fused0-host-explicit

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

325/651/0

109337538.0

111234789.5

1.0173522427402746

0.9934107932225871

1.0429872504429158

supported_no_material_regression

115685029.5 | 111886248.0 | 107283020.0 | 105756308.5 | 110241012.0 | 110101347.0 | 109040920.0 | 109813717.0

114150209.0 | 114365888.5 | 99425894.0 | 108641265.5 | 109970779.5 | 111244084.5 | 109169274.0 | 112407659.0

poisson-128

torch-cuda-eager

325/651/0

186551963.0

195697878.5

1.0490261016443982

0.9985410694352075

1.100851791524981

inconclusive

202371292.5 | 191932768.0 | 180487445.5 | 175281829.5 | 192029493.0 | 180350702.5 | 191600815.5 | 188323792.5

210258255.5 | 205511902.0 | 183410114.0 | 184364183.5 | 191759220.0 | 192941364.5 | 197010004.5 | 199644672.0

poisson-32

cuda-native

92/185/0

29144396.5

28816206.5

0.9887391732403861

0.9411716012051092

1.023246173327046

supported_no_material_regression

29701111.0 | 32504556.5 | 28091494.5 | 28506860.5 | 29179148.5 | 28116036.0 | 30943607.0 | 28152868.5

29239219.5 | 28656043.0 | 28961127.5 | 27319850.0 | 27464350.5 | 30716656.5 | 28921798.0 | 28955217.0

poisson-32

torch-cuda-eager

92/185/0

53020961.0

53376357.0

1.006702933958515

0.9454855351956726

1.0589327423987143

inconclusive

53958199.0 | 59115854.0 | 54770033.5 | 51450306.0 | 53688877.5 | 48385214.0 | 55938665.0 | 51311552.5

53731417.0 | 52970321.0 | 54470296.5 | 49388898.5 | 54153580.5 | 57524409.5 | 53513773.5 | 52594595.5

poisson-64

cuda-native

174/349/0

56961936.0

55889216.0

0.9811677749155155

0.9249600443910291

1.03946636624452

supported_no_material_regression

67347700.0 | 58122404.5 | 54516890.0 | 54653333.0 | 54796247.5 | 52900426.0 | 60597975.5 | 56490412.5

60576613.5 | 56251744.5 | 58827494.0 | 52322990.0 | 52527234.5 | 55822125.5 | 54720425.0 | 56648307.0

poisson-64

torch-cuda-eager

174/349/0

105461594.0

100139081.0

0.9495312672782094

0.899560476056306

1.0183091327823715

supported_no_material_regression

115020169.5 | 101667210.5 | 101360671.5 | 106272526.0 | 104202195.5 | 91747992.0 | 115986158.0 | 103697039.5

104881348.5 | 104283478.0 | 103126335.0 | 93910011.5 | 98000821.5 | 99820982.0 | 97881209.0 | 102225626.5

poisson-8

cuda-native

26/53/0

8191794.0

7692278.5

0.9390224534454846

0.8678008170545711

1.0207222869809929

supported_no_material_regression

8102610.0 | 9878810.5 | 7833115.5 | 7421531.5 | 8439663.5 | 8648737.5 | 8329989.0 | 7259584.5

7747684.5 | 7617019.5 | 8134702.5 | 8176257.5 | 7232534.0 | 7365879.0 | 7496684.0 | 7955650.5

poisson-8

torch-cuda-eager

26/53/0

15271332.5

15548156.0

1.0181270036521044

0.9431097491255604

1.0882743645065225

inconclusive

15103136.0 | 16441349.0 | 15707100.5 | 13671491.0 | 15094727.5 | 15655356.5 | 16063573.5 | 14566101.5

16228801.0 | 15785609.5 | 16078884.5 | 14728729.5 | 15408220.0 | 15106099.5 | 13863019.0 | 15845724.5

cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-explicit

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

325/651/0

60995322.0

60580499.0

0.9931991013999402

0.9533944527717264

1.0540988268106721

inconclusive

59908109.0 | 59706975.5 | 59514904.0 | 60726473.5 | 59091381.5 | 56880350.5 | 64771015.5 | 64123170.0

64833905.5 | 56588012.5 | 58521896.5 | 59686836.5 | 59904873.5 | 65234902.0 | 64112756.0 | 63319699.0

poisson-128

torch-cuda-eager

325/651/0

176377408.5

176029493.0

0.9980274372837268

0.9560784676610117

1.059648760836044

inconclusive

173960215.0 | 177196576.0 | 176462714.0 | 176653276.5 | 176199441.0 | 147917415.0 | 186523638.0 | 186191476.0

191716848.0 | 171008236.0 | 170239259.0 | 174981823.0 | 156390789.0 | 197707006.0 | 190807171.0 | 179998690.5

poisson-32

cuda-native

92/185/0

17332078.5

16537818.5

0.9541739901535756

0.8731887924488706

1.011528253066528

supported_no_material_regression

16813588.0 | 16131604.5 | 17124801.5 | 17003449.5 | 18076274.0 | 22776798.5 | 17767519.0 | 18267824.0

16550236.5 | 16464599.5 | 16472612.0 | 16592352.5 | 14394920.5 | 17305766.0 | 16713023.5 | 19013672.5

poisson-32

torch-cuda-eager

92/185/0

51072423.5

49620923.0

0.971579564850687

0.9037370047039885

1.0335959864506497

supported_no_material_regression

51250257.5 | 48036934.5 | 51195663.0 | 48723898.5 | 51500030.0 | 68507524.5 | 50374487.5 | 50841544.5

48781942.0 | 49542469.5 | 49777746.5 | 49410074.5 | 42923777.0 | 51155321.0 | 54487326.0 | 57152895.0

poisson-64

cuda-native

174/349/0

32344758.5

32774434.5

1.0132842543869975

0.9478825329890102

1.072426589869741

inconclusive

32588408.5 | 30414706.5 | 33320923.5 | 32401356.0 | 31879400.5 | 27130867.0 | 34124444.5 | 33252240.5

32530575.5 | 32380388.0 | 31042345.0 | 33419329.5 | 27730782.5 | 33750865.5 | 40958717.5 | 34216974.0

poisson-64

torch-cuda-eager

174/349/0

95478756.5

97769661.0

1.023993866111987

0.9578466798399796

1.0979202627574849

inconclusive

96864149.5 | 93739385.0 | 101624746.5 | 93385793.0 | 94586831.5 | 79884919.0 | 99579172.5 | 94996668.0

94337237.0 | 102253359.0 | 92970425.0 | 97017106.0 | 80260355.5 | 99079258.5 | 123040599.5 | 99964342.0

poisson-8

cuda-native

26/53/0

4859979.5

4281470.0

0.8809646213528267

0.7992999020910521

0.97470490546205

supported_no_material_regression

4806986.5 | 4721883.0 | 4787934.0 | 4850776.0 | 4842099.0 | 6020817.0 | 5099191.0 | 4248858.0

4671801.5 | 4362797.0 | 4779729.5 | 4520048.5 | 3989491.5 | 4201083.0 | 4124086.5 | 4658392.5

poisson-8

torch-cuda-eager

26/53/0

14898756.5

13374607.0

0.8976995496234871

0.7833976081204889

1.0235781980043934

supported_no_material_regression

14825726.0 | 14077715.5 | 13962275.5 | 14173242.0 | 15921209.0 | 22851177.5 | 14328129.0 | 14751978.0

15069511.0 | 13512552.5 | 14534700.0 | 15210178.5 | 11586062.5 | 11866879.5 | 12466058.5 | 15907986.0

cuda-release-cuda-csr-f64-v1-none-prepared-fused0-device-explicit

Original 5% comparator rows

Case

Implementation

I/M/R

A pooled ns

B pooled ns

B/A

CI lower

CI upper

Verdict

A 8 block medians ns

B 8 block medians ns

poisson-128

cuda-native

325/651/0

65434534.5

64877491.0

0.991487010578489

0.9608346775770368

1.0255285286483091

supported_no_material_regression

64541884.0 | 64067497.5 | 65892902.5 | 63244306.5 | 68693480.0 | 66810175.0 | 64356718.5 | 64154544.5

67326218.5 | 63926808.5 | 66690706.0 | 65479912.0 | 62118783.0 | 61933728.0 | 66164473.0 | 65078927.0

poisson-128

torch-cuda-eager

325/651/0

176463510.5

175011825.0

0.9917734522231438

0.9566215627994861

1.0269680402768757

supported_no_material_regression

175537605.0 | 175642070.0 | 178540381.5 | 174779391.5 | 188208598.5 | 179910421.5 | 171458526.5 | 173049901.5

183800960.5 | 171231376.0 | 184653588.0 | 179312300.5 | 167152306.5 | 167831910.0 | 173342422.5 | 176034510.0

poisson-32

cuda-native

92/185/0

19839304.5

19456317.5

0.980695543031763

0.9434495458893825

1.0270082370836344

supported_no_material_regression

20083342.5 | 20713124.5 | 19835931.0 | 20465563.0 | 20402354.5 | 18868107.0 | 18908994.5 | 19861047.5

19214742.5 | 18822895.0 | 19466560.0 | 20907044.0 | 18716626.5 | 20123885.5 | 18695418.0 | 19700358.5

poisson-32

torch-cuda-eager

92/185/0

51157889.0

51121657.0

0.9992917612374506

0.9381387604400526

1.0534587184014825

inconclusive

52115957.0 | 56912918.0 | 50938055.5 | 54341253.5 | 53490935.0 | 47674445.5 | 47590522.0 | 50492476.5

52135629.5 | 47280815.0 | 52579225.5 | 55499498.0 | 49566360.5 | 56109010.5 | 48809500.0 | 50177660.0

poisson-64

cuda-native

174/349/0

38654419.0

37976335.5

0.9824578012671721

0.9436207784243157

1.036316724000773

supported_no_material_regression

39452965.0 | 41049690.0 | 39122820.5 | 39064379.5 | 39106426.0 | 36789968.5 | 36503664.5 | 38357676.5

37184771.5 | 37020864.0 | 40181407.5 | 41575016.0 | 36640478.5 | 38287094.0 | 37056535.0 | 37340048.0

poisson-64

torch-cuda-eager

174/349/0

99725767.5

98867698.0

0.991395709238337

0.9533958336888976

1.049647381032783

supported_no_material_regression

104153532.0 | 105951501.5 | 98940035.5 | 98563610.5 | 101990198.0 | 96437583.5 | 92539061.0 | 99534427.5

96944045.5 | 94171264.5 | 105137009.0 | 109760610.5 | 94795915.0 | 97540511.0 | 97313611.0 | 98223945.0

poisson-8

cuda-native

26/53/0

5823935.0

5497203.5

0.943898498180354

0.8692980418047835

1.027405753430701

supported_no_material_regression

5413947.0 | 6387084.5 | 5786081.0 | 6269184.0 | 5781243.5 | 5141272.5 | 5198873.5 | 5362605.0

5339402.5 | 5648116.5 | 5697530.0 | 5581641.0 | 4853099.5 | 5931847.5 | 4836919.5 | 5678181.0

poisson-8

torch-cuda-eager

26/53/0

15345315.0

14724162.5

0.9595216846314331

0.9062666617667926

1.0234559853609362

supported_no_material_regression

15230314.5 | 16032934.0 | 15224179.0 | 17272814.5 | 15196708.5 | 14644635.5 | 14586808.0 | 14318622.0

13827690.0 | 15621395.0 | 15661985.0 | 14948081.5 | 14270940.0 | 14969318.0 | 14212865.0 | 14824872.5

Frozen provenance and reproducibility

The companion JSON retains every original comparator report and exact argv, every raw artifact/log digest through the completed manifest, full stage/source/compiler-unit audits, semantic runtime controls, all fixture/work identities, preflight lineage, safety/trace identities, calibration outcomes and this renderer’s digest. Active production differs only in conjugate_gradient.cu. Known inactive baseline-only generated legacy outputs remain in the full source record and are explicitly classified; compiler arguments do not reference them. This is an explicit compiler-unit/argument audit, not a complete transitive-include or linker-input inventory; root separately verified tracked production/build inputs.

Evidence identities

Role

Path

SHA256

renderer

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/render_cuda_terminal_results.py

5cc876f04da046979150f011a96d4356d07abfc1df2ea21e1c17e7d3e08350b7

campaign_plan

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/comparison-cuda-release/plan.json

bb7e83ca0d122dac443d3d70d692e2a3888adf91abbb65815a2931a2962c6548

campaign_manifest

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/comparison-cuda-release/manifest.json

7260ac305c97182602acaeb11ff638722f75abb7f77df0209228ea956e830a00

campaign_execution_claim

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/comparison-cuda-release/execution-claim.json

4149f55a771de42460469deab00989e426a36efc7f41278dfaa6884fb91ee061

frozen_runner

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/tools-v2/run_performance_campaign.py

f98dc4259cb8c142c6661a273c5b18de16f75c5f7b141ef8f1fce0d872ac8b33

original_comparator

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/tools-v2/compare_performance.py

b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063

preflight plan

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/preflight-cuda-v1/plan.json

db995bca8e50640866a93241a8d300f6a80ada173836f1bf34ba3578baf6491a

preflight manifest

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/preflight-cuda-v1/manifest.json

18944e6b6e9684f0a67b9b22ae8b131c3ce19b281fcb5c7b966795391ff93fba

preflight execution_claim

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/preflight-cuda-v1/execution-claim.json

383bbbfbe352ce2fc2847fcaf57fcef6704c980584eb272ffbf8845964f9fcec

diagnostics analysis

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/cuda-terminal-analysis.json

a9933b0645a97c85c688788a36fed0501210714802e7703999e96f8ac0cd6a20

diagnostics report

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/cuda-terminal-analysis.rst

b92a8a4cc264539e30859c3cd039063eea471124f07e0012303c14fd8b19846e

diagnostics plan

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/cuda-terminal-diagnostics/plan.json

26a5b5a6f273da0806adc77cf06e49b0f7e1d5cbabdf7770f0260c8a3892c2ac

diagnostics manifest

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/cuda-terminal-diagnostics/manifest.json

5407dea8a0e690e415b49e5633b01ff25f6ac087f3f74b8fa63bb7849eda156c

Actual measured native modules

Source

Module

Path

SHA256

baseline

_core

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/baseline-stages/cuda-release/python/cgpy/_core.cpython-312-x86_64-linux-gnu.so

84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a

baseline

_cuda

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/baseline-stages/cuda-release/python/cgpy/_cuda.cpython-312-x86_64-linux-gnu.so

2b95332f5e7cfa622e5de9eeb109ed8d6b2798f7e9eb9f8ee7d686a9c1f71fbe

current

_core

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/cuda-terminal-stages/cuda-release/python/cgpy/_core.cpython-312-x86_64-linux-gnu.so

84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a

current

_cuda

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/cuda-terminal-stages/cuda-release/python/cgpy/_cuda.cpython-312-x86_64-linux-gnu.so

7351610ce02c04589482fcd8dea1a16232ce060a93ae087d043d4c99127a317e

Exact timed fixture identities

Fixture spec

Input SHA256

{“condition_scale”: 100.0, “grid”: 128, “name”: “poisson-128”, “size”: 16384, “storage”: “csr”}

2eb8d7cd2473ae1f9040400f2865f064a7b38993dce3315c082956147070f2f2

{“condition_scale”: 100.0, “grid”: 32, “name”: “poisson-32”, “size”: 1024, “storage”: “csr”}

c0eebda4437f2242475d0873f2614220a7a10a33f3806cc61eb295855c313da5

{“condition_scale”: 100.0, “grid”: 64, “name”: “poisson-64”, “size”: 4096, “storage”: “csr”}

0e3ab2d319ca119fa0519a6a64fa216b257bfb6624207c782cae586b010e5fa5

{“condition_scale”: 100.0, “grid”: 8, “name”: “poisson-8”, “size”: 64, “storage”: “csr”}

8a5ea77fe69144781dcbcb30e2421cb604751f85bf97bbb97f3e3770c9b18f4e

{“condition_scale”: 100.0, “name”: “dense-1024”, “size”: 1024}

204aa61929e5e8d34102ac2d0bda5714b6f0d17f8bf4796a7f71b535a21053bb

{“condition_scale”: 100.0, “name”: “dense-2048”, “size”: 2048}

8bda02fed9ccafa8c12f857ad51962b02d8c860ee96a006078aed872c40cae6c

{“condition_scale”: 100.0, “name”: “dense-256”, “size”: 256}

6a6622788ec7f29156523e562995e09db8ec35032185f8c76ae6fccf2165a4a0

{“condition_scale”: 100.0, “name”: “dense-64”, “size”: 64}

b2e274e240704161b83bff5de50cf7b8ba9e3848b098f027c1ba8b5fccf82a29

Retained A/A calibration context

Manifest

SHA256

Original verdict

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/calibration-cpu-release/manifest.json

9635ade929760cd3787901edfd097782f5fbc92bc3e9beef490f87ea7be76ab9

inconclusive

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/calibration-cpu-blas/manifest.json

5c8f114588219f50c619bb6619c12be28257e4909c0b775818cbf1f1eff849bf

inconclusive

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/calibration-cpu-openmp/manifest.json

0e226a0c3d9924e46545fcdc4349069b2e284663c10f1b21492e7238e141cf95

inconclusive

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/calibration-cuda-release/manifest.json

f22ee7194333605172a49abe73d50bfea245d38ba3909a099c370b662ad9516f

inconclusive

Audit command (run only after all serialized measurements finish):

/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/.venv-cuda/bin/python /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/render_cuda_terminal_results.py --campaign-dir /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/comparison-cuda-release --preflight-dir /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/preflight-cuda-v1 --output-prefix /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/reliable-performance-20261018/cuda-terminal-results

This final audit additionally rehashes the actual compiler-unit/test-source paths and diagnostic test binaries declared during preparation. Execute it before those live source paths are edited or their worktrees/builds are archived. A later replay requires preserving or restoring those exact paths and bytes; the frozen source declarations and already emitted report retain the historical identities. Reading the final report does not require those live paths.

The completed campaign cannot resume or rerun into these paths. A new measurement requires a newly frozen versioned evidence directory; the present raw artifacts and comparator reports remain unchanged.

Limits

These intervals describe this measured host and these supplied blocks. Eight processes are not eight independent physical-host replications. A 60-second boundary pause and boundary telemetry do not establish thermal stability or absence of WSL/background contention. Within-block serial dependence can make round bootstrap intervals too narrow; per-row 95% intervals do not establish simultaneous family-wise coverage. Earlier A/A and A/B outcomes remain disclosed and unpooled. Controls reveal context but cannot rescue a failed native row, and a changing control does not by itself identify the cause of a native change. No unsupported kernel-only, allocation-only, scope-difference or universal NumPy/PyTorch speedup claim follows from this campaign.