Last active
July 25, 2023 20:09
-
-
Save HDCharles/3deb02e26e3d53f106c15216ff0e608a to your computer and use it in GitHub Desktop.
benchmarks for mixed_precision triton kernel
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| M N K Abf16Wint8 cublas_bf16_linear ratio (higher is good) | |
| (5120, 1280, 1280) 0.1783519983291626 0.09504000097513199 0.5328788119308213 | |
| (5120, 1280, 3840) 0.41705599427223206 0.2657119929790497 0.6371134730786461 | |
| (5120, 1280, 5120) 0.548255980014801 0.39129599928855896 0.7137104081892464 | |
| (5120, 5120, 1280) 0.6700800061225891 0.3110080063343048 0.46413563080914677 | |
| (5120, 5120, 3840) 1.6803359985351562 0.9334560036659241 0.5555174706009217 | |
| (5120, 5120, 5120) 2.511807918548584 1.164560079574585 0.46363420983540454 | |
| (16384, 1280, 1280) 0.48159998655319214 0.28620800375938416 0.5942857386848636 | |
| (16384, 1280, 3840) 1.2650560140609741 0.7705600261688232 0.6091113892223932 | |
| (16384, 1280, 5120) 1.7240959405899048 1.2534880638122559 0.727040783695234 | |
| (16384, 5120, 1280) 1.8544960021972656 0.9801920056343079 0.5285489990126403 | |
| (16384, 5120, 3840) 5.158592224121094 2.9050240516662598 0.5631427966107961 | |
| (16384, 5120, 5120) 6.766384124755859 3.9079999923706055 0.5775610607255631 | |
| (19600, 1280, 1280) 0.5773760080337524 0.34537601470947266 0.5981821376431051 | |
| (19600, 1280, 3840) 1.5184320211410522 0.8964160084724426 0.5903563649815651 | |
| (19600, 1280, 5120) 2.0843679904937744 1.501471996307373 0.720348807482734 | |
| (19600, 5120, 1280) 2.2641279697418213 1.2023520469665527 0.5310442090884364 | |
| (19600, 5120, 3840) 6.234176158905029 3.5112640857696533 0.563228243198436 | |
| (19600, 5120, 5120) 8.173471450805664 4.711040019989014 0.5763817795588732 | |
| (10240, 1280, 1280) 0.3123840093612671 0.17897599935531616 0.5729358545633279 | |
| (10240, 1280, 3840) 0.8005279898643494 0.4816640019416809 0.6016828993365986 | |
| (10240, 1280, 5120) 1.0743680000305176 0.791808009147644 0.7369988766653071 | |
| (10240, 5120, 1280) 1.1649919748306274 0.6250240206718445 0.5365050010432164 | |
| (10240, 5120, 3840) 3.2482879161834717 1.874079942703247 0.5769439135509792 | |
| (10240, 5120, 5120) 4.261663913726807 2.4463040828704834 0.5740255760176078 | |
| (32768, 1280, 1280) 0.9344639778137207 0.5667040348052979 0.6064482401249571 | |
| (32768, 1280, 3840) 2.5094079971313477 1.5450559854507446 0.6157053724292699 | |
| (32768, 1280, 5120) 3.449984073638916 2.521120071411133 0.7307628144358209 | |
| (32768, 5120, 1280) 3.72214412689209 1.9629759788513184 0.5273777457108736 | |
| (32768, 5120, 3840) 10.300543785095215 5.878496170043945 0.5706976537054356 | |
| (32768, 5120, 5120) 13.574848175048828 7.7315521240234375 0.569549804485797 | |
| (39200, 1280, 1280) 1.118399977684021 0.6743999719619751 0.6030042788077664 | |
| (39200, 1280, 3840) 2.9890880584716797 1.868607997894287 0.6251431745539494 | |
| (39200, 1280, 5120) 4.123472213745117 3.016767978668213 0.731608659472026 | |
| (39200, 5120, 1280) 4.448160171508789 2.343247890472412 0.5267903582881991 | |
| (39200, 5120, 3840) 12.371456146240234 7.027520179748535 0.5680430902132927 | |
| (39200, 5120, 5120) 16.269119262695312 9.447664260864258 0.580711476037152 | |
| (20480, 1280, 1280) 0.5941759943962097 0.3586080074310303 0.6035383637392502 | |
| (20480, 1280, 3840) 1.5621119737625122 0.9106400012969971 0.5829543698481641 | |
| (20480, 1280, 5120) 2.150207996368408 1.4994560480117798 0.6973539539171488 | |
| (20480, 5120, 1280) 2.3274879455566406 1.2096960544586182 0.519743209312007 | |
| (20480, 5120, 3840) 6.431935787200928 3.6903839111328125 0.573759445558583 | |
| (20480, 5120, 5120) 8.511455535888672 4.877200126647949 0.57301599075307 | |
| (65536, 1280, 1280) 1.849951982498169 1.1407359838485718 0.6166300502071009 | |
| (65536, 1280, 3840) 4.946096420288086 3.1222400665283203 0.631253376646965 | |
| (65536, 1280, 5120) 6.8432159423828125 5.154895782470703 0.7532855642541312 | |
| (65536, 5120, 1280) 7.474368095397949 3.9599838256835938 0.5298085102500905 | |
| (65536, 5120, 3840) 20.491985321044922 11.796127319335938 0.5756458993371186 | |
| (65536, 5120, 5120) 27.292320251464844 15.615535736083984 0.5721585996429109 | |
| (78400, 1280, 1280) 2.2279839515686035 1.380031943321228 0.6194083859309767 | |
| (78400, 1280, 3840) 5.925439834594727 3.806623935699463 0.6424204855604307 | |
| (78400, 1280, 5120) 8.250384330749512 6.113183975219727 0.7409574790880524 | |
| (78400, 5120, 1280) 8.912223815917969 4.63270378112793 0.5198145689354814 | |
| (78400, 5120, 3840) 24.794078826904297 14.050623893737793 0.5666927169115605 | |
| (78400, 5120, 5120) 32.753055572509766 18.722911834716797 0.5716386305780666 | |
| (40960, 1280, 1280) 1.172368049621582 0.7157760262489319 0.6105386669996429 | |
| (40960, 1280, 3840) 3.109855890274048 1.939743995666504 0.6237407983221914 | |
| (40960, 1280, 5120) 4.274335861206055 3.1879360675811768 0.7458319072478461 | |
| (40960, 5120, 1280) 4.6715521812438965 2.4270079135894775 0.5195292312764528 | |
| (40960, 5120, 3840) 12.891327857971191 7.334639549255371 0.5689591972265362 | |
| (40960, 5120, 5120) 18.869728088378906 9.784480094909668 0.5185278796325384 | |
| (131072, 1280, 1280) 3.688591957092285 2.3401920795440674 0.6344404875265302 | |
| (131072, 1280, 3840) 9.862735748291016 7.187808036804199 0.7287844083270385 | |
| (131072, 1280, 5120) 13.671615600585938 10.169631958007812 0.7438500507264089 | |
| (131072, 5120, 1280) 14.878560066223145 7.86953592300415 0.5289178447361538 | |
| (131072, 5120, 3840) 40.95172882080078 23.393936157226562 0.5712563750261987 | |
| (131072, 5120, 5120) 54.15119934082031 31.183744430541992 0.5758643356036444 | |
| (156800, 1280, 1280) 4.428383827209473 2.745039939880371 0.6198739872126546 | |
| (156800, 1280, 3840) 11.834848403930664 7.428800106048584 0.6277055567168299 | |
| (156800, 1280, 5120) 16.31332778930664 12.164527893066406 0.7456803449410386 | |
| (156800, 5120, 1280) 17.776256561279297 9.382207870483398 0.5277943552479979 | |
| (156800, 5120, 3840) 49.37774658203125 27.723487854003906 0.5614571294367773 | |
| (156800, 5120, 5120) 65.15977478027344 36.47092819213867 0.5597153813855712 | |
| (81920, 1280, 1280) 2.3414719104766846 1.450943946838379 0.619671728858362 | |
| (81920, 1280, 3840) 6.149263858795166 3.9095680713653564 0.635778226652867 | |
| (81920, 1280, 5120) 8.624799728393555 6.346831798553467 0.7358816434495483 | |
| (81920, 5120, 1280) 9.345855712890625 4.882976055145264 0.5224750097960783 | |
| (81920, 5120, 3840) 25.97123146057129 14.669856071472168 0.5648502302920632 | |
| (81920, 5120, 5120) 34.00536346435547 19.529983520507812 0.5743206815295241 | |
| (262144, 1280, 1280) 7.414432048797607 4.625040054321289 0.6237888517801344 | |
| (262144, 1280, 3840) 19.695999145507812 14.543695449829102 0.7384086149874844 | |
| (262144, 1280, 5120) 27.57036781311035 20.22449493408203 0.7335591266383038 | |
| (262144, 5120, 1280) 29.861120223999023 15.594911575317383 0.5222480422145697 | |
| (262144, 5120, 3840) 82.37635040283203 47.06662368774414 0.5713608755131986 | |
| (262144, 5120, 5120) 112.52588653564453 61.04915237426758 0.5425342936972036 | |
| (313600, 1280, 1280) 8.87548828125 5.600576400756836 0.6310161450596908 | |
| (313600, 1280, 3840) 23.94443130493164 18.393951416015625 0.7681932881081696 | |
| (313600, 1280, 5120) 32.954654693603516 24.502016067504883 0.7435069884759167 | |
| (313600, 5120, 1280) 35.742576599121094 18.70172882080078 0.5232339299584982 | |
| (313600, 5120, 3840) 99.6239013671875 56.35862350463867 0.5657138772041822 | |
| (313600, 5120, 5120) 131.56675720214844 73.92972564697266 0.5619179739558513 |
Author
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
newest version of triton
(5120, 1280, 1280) 0.1223360002040863 0.09520000219345093 0.7781846883553009
(5120, 1280, 3840) 0.30318400263786316 0.26492801308631897 0.8738192344625818
(5120, 1280, 5120) 0.3832319974899292 0.3914079964160919 1.0213343326750204
(5120, 5120, 1280) 0.36934399604797363 0.31036800146102905 0.8403223141082703
(5120, 5120, 3840) 1.032639980316162 0.933247983455658 0.9037496138488911
(5120, 5120, 5120) 1.3103359937667847 1.2100319862365723 0.9234516887215535
(16384, 1280, 1280) 0.3473919928073883 0.28652799129486084 0.8247973391077165
(16384, 1280, 3840) 0.9122079610824585 0.7712960243225098 0.845526521613846
(16384, 1280, 5120) 1.2064319849014282 1.2750719785690308 1.056895038034996
(16384, 5120, 1280) 1.1163519620895386 0.9713919758796692 0.8701484915755963
(16384, 5120, 3840) 3.1203200817108154 2.9056639671325684 0.9312070207680249
(16384, 5120, 5120) 4.161231994628906 3.9087040424346924 0.9393141376111297
(19600, 1280, 1280) 0.4209280014038086 0.34409600496292114 0.8174699801755877
(19600, 1280, 3840) 1.1051840782165527 0.8720960021018982 0.7890956984371407
(19600, 1280, 5120) 1.447424054145813 1.5262079238891602 1.0544303996590971
(19600, 5120, 1280) 1.4601919651031494 1.202847957611084 0.8237601537042519
(19600, 5120, 3840) 3.812448024749756 3.578752040863037 0.9387018570824822
(19600, 5120, 5120) 5.061504364013672 4.710720062255859 0.9306956437197167
(10240, 1280, 1280) 0.23097600042819977 0.1823039948940277 0.7892767844107594
(10240, 1280, 3840) 0.5777919888496399 0.4851199984550476 0.8396101154342787
(10240, 1280, 5120) 0.7522720098495483 0.7988640069961548 1.0619350401670862
(10240, 5120, 1280) 0.7792160511016846 0.6329120397567749 0.812242046171842
(10240, 5120, 3840) 1.999216079711914 1.8749120235443115 0.9378236012459871
(10240, 5120, 5120) 2.5550239086151123 2.472543954849243 0.9677185197806719
(32768, 1280, 1280) 0.6780800223350525 0.5750719904899597 0.8480886791349908
(32768, 1280, 3840) 1.833359956741333 1.5610560178756714 0.8514727356925246
(32768, 1280, 5120) 2.4307520389556885 2.5569279193878174 1.0519081660366876
(32768, 5120, 1280) 2.277423858642578 1.9847359657287598 0.8714829074073764
(32768, 5120, 3840) 6.37772798538208 5.920864105224609 0.9283657313067264
(32768, 5120, 5120) 8.313183784484863 7.82528018951416 0.9413096585352423
(39200, 1280, 1280) 0.8366720080375671 0.6808000206947327 0.8137000092683444
(39200, 1280, 3840) 2.199968099594116 1.8853440284729004 0.8569869848661612
(39200, 1280, 5120) 2.9179840087890625 3.058687925338745 1.048219563961241
(39200, 5120, 1280) 2.8849599361419678 2.37174391746521 0.8221063619472383
(39200, 5120, 3840) 7.720448017120361 7.118880271911621 0.9220812388251637
(39200, 5120, 5120) 10.026575088500977 9.447343826293945 0.942230397010508
(20480, 1280, 1280) 0.43462398648262024 0.3681600093841553 0.8470770616312435
(20480, 1280, 3840) 1.15721595287323 0.9756159782409668 0.8430716633473883
(20480, 1280, 5120) 1.4956799745559692 1.5578560829162598 1.0415704625441342
(20480, 5120, 1280) 1.4897760152816772 1.2235519886016846 0.8212992933507144
(20480, 5120, 3840) 3.9598400592803955 3.735167980194092 0.9432623349117963
(20480, 5120, 5120) 5.230912208557129 4.967264175415039 0.9495980772319608
(65536, 1280, 1280) 1.3661760091781616 1.1707199811935425 0.8569320302277904
(65536, 1280, 3840) 3.6805760860443115 3.183295965194702 0.8648906830821531
(65536, 1280, 5120) 5.01910400390625 5.158623695373535 1.0277977287098854
(65536, 5120, 1280) 4.675327777862549 3.9884800910949707 0.8530910089299474
(65536, 5120, 3840) 12.292736053466797 11.892816543579102 0.9674670058684852
(65536, 5120, 5120) 16.058223724365234 15.813360214233398 0.984751519574341
(78400, 1280, 1280) 1.6692160367965698 1.3972480297088623 0.8370684194901173
(78400, 1280, 3840) 4.464240074157715 3.803391933441162 0.8519685030959636
(78400, 1280, 5120) 5.83462381362915 6.172783851623535 1.0579574705749621
(78400, 5120, 1280) 5.63756799697876 4.708896160125732 0.8352708406620171
(78400, 5120, 3840) 14.677791595458984 14.445759773254395 0.9841916394100884
(78400, 5120, 5120) 18.93075180053711 18.965408325195312 1.0018306998593274
(40960, 1280, 1280) 0.8642240166664124 0.7280640006065369 0.8424482386116876
(40960, 1280, 3840) 2.321983814239502 1.971695899963379 0.8491428268672709
(40960, 1280, 5120) 3.0811519622802734 3.2201919555664062 1.0451259772280863
(40960, 5120, 1280) 2.8427679538726807 2.5039520263671875 0.8808147787638007
(40960, 5120, 3840) 7.896543979644775 7.473152160644531 0.9463826428255655
(40960, 5120, 5120) 10.458784103393555 9.835583686828613 0.9404136838083567
(131072, 1280, 1280) 2.752511978149414 2.3405439853668213 0.8503301725649275
(131072, 1280, 3840) 7.472288131713867 7.281120300292969 0.9744164266619291
(131072, 1280, 5120) 9.988592147827148 10.248127937316895 1.0259832202225019
(131072, 5120, 1280) 9.38049602508545 7.977343559265137 0.850418094942103
(131072, 5120, 3840) 24.614944458007812 23.46267318725586 0.9531881425644618
(131072, 5120, 5120) 32.49158477783203 31.737472534179688 0.9767905367248553
(156800, 1280, 1280) 3.2626559734344482 2.826575994491577 0.8663420285517172
(156800, 1280, 3840) 8.933919906616211 7.628096103668213 0.8538352910483402
(156800, 1280, 5120) 11.939136505126953 12.272704124450684 1.0279390070781491
(156800, 5120, 1280) 11.188464164733887 9.501520156860352 0.849224702958714
(156800, 5120, 3840) 29.652416229248047 28.46224021911621 0.9598624273674572
(156800, 5120, 5120) 39.12627410888672 36.947662353515625 0.944318445725036
(81920, 1280, 1280) 1.706496000289917 1.4764480590820312 0.8651928037517798
(81920, 1280, 3840) 4.690720081329346 4.019023895263672 0.8568031827907933
(81920, 1280, 5120) 6.135056018829346 6.4370880126953125 1.0492305193202782
(81920, 5120, 1280) 5.866911888122559 5.005375862121582 0.8531534063524734
(81920, 5120, 3840) 15.21183967590332 14.820943832397461 0.9743031841096073
(81920, 5120, 5120) 20.182239532470703 19.89481544494629 0.9857585632624177
(262144, 1280, 1280) 5.471663951873779 4.703904151916504 0.8596844019095224
(262144, 1280, 3840) 15.177103996276855 14.598272323608398 0.9618615202998904
(262144, 1280, 5120) 19.359615325927734 20.750015258789062 1.0718196053719728
(262144, 5120, 1280) 18.73401641845703 15.782400131225586 0.8424461566968913
(262144, 5120, 3840) 49.89020538330078 47.46022415161133 0.9512934209626883
(262144, 5120, 5120) 65.15248107910156 61.88032150268555 0.9497768999396463
(313600, 1280, 1280) 6.589119911193848 5.630847930908203 0.8545675305350422
(313600, 1280, 3840) 17.812063217163086 18.60857582092285 1.0447175935796293
(313600, 1280, 5120) 23.498672485351562 24.78575897216797 1.054772731847671
(313600, 5120, 1280) 22.920223236083984 18.866304397583008 0.8231291730126444
(313600, 5120, 3840) 59.40399932861328 56.791038513183594 0.9560137222247409
(313600, 5120, 5120) 77.88349151611328 73.92371368408203 0.9491576744320456