{"id":14806,"date":"2022-10-09T05:49:57","date_gmt":"2022-10-09T00:19:57","guid":{"rendered":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/"},"modified":"2022-10-09T05:49:57","modified_gmt":"2022-10-09T00:19:57","slug":"solved-nvidia-tesla-t4-tensor-core-benchmark-closed","status":"publish","type":"post","link":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/","title":{"rendered":"[Solved] Nvidia Tesla T4 tensor core benchmark [closed]"},"content":{"rendered":"<p> [ad_1]<br \/>\n<\/p>\n<div id=\"answer-65751263\" class=\"answer js-answer accepted-answer js-accepted-answer\" data-answerid=\"65751263\" data-parentid=\"65741716\" data-score=\"1\" data-position-on-page=\"1\" data-highest-scored=\"1\" data-question-has-accepted-highest-score=\"1\" itemprop=\"acceptedAnswer\" itemscope itemtype=\"https:\/\/schema.org\/Answer\">\n<div class=\"post-layout\">\n<div class=\"votecell post-layout--left\"><\/div>\n<div class=\"answercell post-layout--right\">\n<div class=\"s-prose js-post-body\" itemprop=\"text\">\n<p>This might be more of an extended comment, bet hear me out &#8230;<\/p>\n<p>As pointed out in the comments CUDA Samples are not meant as performance measuring tools.<br \/>\nThe second benchmark you provided does not actually use tensor cores, but just a normal instruction executed on FP32 or FP64 cores.<\/p>\n<pre><code>for(int i=0; i&lt;compute_iterations; i++){\n            tmps[j] = mad(tmps[j], tmps[j], seed);\n    }\n<\/code><\/pre>\n<p>On a Turing T4 this, for single precision operations gives me a peak of 7.97 TFLOPS, so very close to the theoretical limit of 8.1 TFLOPS.<br \/>\nFor half precision operations I get 16.09 TFLOPS, as expected about double that of the single precision performance.<\/p>\n<p>Now, on to Tensor cores. As the previously mentioned benchmark does not use them, let&#8217;s look for something that does.<br \/>\nCUTLASS (<a rel=\"nofollow noopener\" target=\"_blank\" href=\"https:\/\/github.com\/NVIDIA\/cutlass\">https:\/\/github.com\/NVIDIA\/cutlass<\/a>) is a high performance Matrix-Matrix Multiplication library from NVIDIA.<br \/>\nThey provide a profiling application for all the kernels provided. If you run this on a T4, you should get output like this:<\/p>\n<pre><code>Problem ID: 1\n\n   Provider: ^[[1;37mCUTLASS^[[0m\n   OperationKind: ^[[1;37mgemm^[[0m\n   Operation: cutlass_tensorop_h1688gemm_256x128_32x2_nt_align8\n\n      Status: ^[[1;37mSuccess^[[0m\nVerification: ^[[1;37mON^[[0m\n Disposition: ^[[1;32mPassed^[[0m\n\n reference_device: Passed\n      cuBLAS: Passed\n\n   Arguments: --gemm_kind=universal --m=1024 --n=1024 --k=1024 --A=f16:column --B=f16:row --C=f16:column --alpha=1  \\\n              --beta=0 --split_k_slices=1 --batch_count=1 --op_class=tensorop --accum=f16 --cta_m=256 --cta_n=128  \\\n              --cta_k=32 --stages=2 --warps_m=4 --warps_n=2 --warps_k=1 --inst_m=16 --inst_n=8 --inst_k=8 --min_cc=75  \\\n              --max_cc=1024\n\n       Bytes: 6291456  bytes\n       FLOPs: 2149580800  flops\n\n     Runtime: 0.0640419  ms\n      Memory: 91.4928 GiB\/s\n\n        Math: 33565.2 GFLOP\/s\n<\/code><\/pre>\n<p>As you can see we are now actually using Tensor cores, and half-precision operation, with a performance of 33.5 TFLOPS. Now, this might not be at 65 TFLOS, but for an application you can use in the real world, that is pretty good.<\/p>\n<\/p><\/div>\n<div class=\"mt24\"><\/div>\n<\/div>\n<p>            <span class=\"d-none\" itemprop=\"commentCount\">3<\/span> <\/p><\/div>\n<\/div>\n<p>[ad_2]<\/p>\n<p>solved Nvidia Tesla T4 tensor core benchmark [closed] <\/p>\n","protected":false},"excerpt":{"rendered":"<p>[ad_1] This might be more of an extended comment, bet hear me out &#8230; As pointed out in the comments CUDA Samples are not meant as performance measuring tools. The second benchmark you provided does not actually use tensor cores, but just a normal instruction executed on FP32 or FP64 cores. for(int i=0; i&lt;compute_iterations; i++){ &#8230; <a title=\"[Solved] Nvidia Tesla T4 tensor core benchmark [closed]\" class=\"read-more\" href=\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/\" aria-label=\"More on [Solved] Nvidia Tesla T4 tensor core benchmark [closed]\">Read more<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[320],"tags":[3226,1564,1644,3302],"class_list":["post-14806","post","type-post","status-publish","format-standard","hentry","category-solved","tag-benchmarking","tag-cuda","tag-gpu","tag-nvidia"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v26.5 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>[Solved] Nvidia Tesla T4 tensor core benchmark [closed] - JassWeb<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"[Solved] Nvidia Tesla T4 tensor core benchmark [closed] - JassWeb\" \/>\n<meta property=\"og:description\" content=\"[ad_1] This might be more of an extended comment, bet hear me out &#8230; As pointed out in the comments CUDA Samples are not meant as performance measuring tools. The second benchmark you provided does not actually use tensor cores, but just a normal instruction executed on FP32 or FP64 cores. for(int i=0; i&lt;compute_iterations; i++){ ... Read more\" \/>\n<meta property=\"og:url\" content=\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/\" \/>\n<meta property=\"og:site_name\" content=\"JassWeb\" \/>\n<meta property=\"article:published_time\" content=\"2022-10-09T00:19:57+00:00\" \/>\n<meta name=\"author\" content=\"Kirat\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Kirat\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"2 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/\"},\"author\":{\"name\":\"Kirat\",\"@id\":\"https:\/\/jassweb.com\/solved\/#\/schema\/person\/65c9c7b7958150c0dc8371fa35dd7c31\"},\"headline\":\"[Solved] Nvidia Tesla T4 tensor core benchmark [closed]\",\"datePublished\":\"2022-10-09T00:19:57+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/\"},\"wordCount\":208,\"publisher\":{\"@id\":\"https:\/\/jassweb.com\/solved\/#organization\"},\"keywords\":[\"benchmarking\",\"cuda\",\"gpu\",\"nvidia\"],\"articleSection\":[\"Solved\"],\"inLanguage\":\"en-US\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/\",\"url\":\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/\",\"name\":\"[Solved] Nvidia Tesla T4 tensor core benchmark [closed] - JassWeb\",\"isPartOf\":{\"@id\":\"https:\/\/jassweb.com\/solved\/#website\"},\"datePublished\":\"2022-10-09T00:19:57+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\/\/jassweb.com\/solved\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"[Solved] Nvidia Tesla T4 tensor core benchmark [closed]\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/jassweb.com\/solved\/#website\",\"url\":\"https:\/\/jassweb.com\/solved\/\",\"name\":\"JassWeb\",\"description\":\"Build High-quality Websites\",\"publisher\":{\"@id\":\"https:\/\/jassweb.com\/solved\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/jassweb.com\/solved\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/jassweb.com\/solved\/#organization\",\"name\":\"Jass Web\",\"url\":\"https:\/\/jassweb.com\/solved\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/jassweb.com\/solved\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/jassweb.com\/wp-content\/uploads\/2021\/02\/jass-website-logo-1.png\",\"contentUrl\":\"https:\/\/jassweb.com\/wp-content\/uploads\/2021\/02\/jass-website-logo-1.png\",\"width\":693,\"height\":132,\"caption\":\"Jass Web\"},\"image\":{\"@id\":\"https:\/\/jassweb.com\/solved\/#\/schema\/logo\/image\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\/\/jassweb.com\/solved\/#\/schema\/person\/65c9c7b7958150c0dc8371fa35dd7c31\",\"name\":\"Kirat\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\/\/jassweb.com\/solved\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/jassweb.com\/solved\/wp-content\/litespeed\/avatar\/1261af3c9451399fa1336d28b98ea3bb.jpg?ver=1776403586\",\"contentUrl\":\"https:\/\/jassweb.com\/solved\/wp-content\/litespeed\/avatar\/1261af3c9451399fa1336d28b98ea3bb.jpg?ver=1776403586\",\"caption\":\"Kirat\"},\"sameAs\":[\"http:\/\/jassweb.com\"],\"url\":\"https:\/\/jassweb.com\/solved\/author\/jaspritsinghghumangmail-com\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"[Solved] Nvidia Tesla T4 tensor core benchmark [closed] - JassWeb","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/","og_locale":"en_US","og_type":"article","og_title":"[Solved] Nvidia Tesla T4 tensor core benchmark [closed] - JassWeb","og_description":"[ad_1] This might be more of an extended comment, bet hear me out &#8230; As pointed out in the comments CUDA Samples are not meant as performance measuring tools. The second benchmark you provided does not actually use tensor cores, but just a normal instruction executed on FP32 or FP64 cores. for(int i=0; i&lt;compute_iterations; i++){ ... Read more","og_url":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/","og_site_name":"JassWeb","article_published_time":"2022-10-09T00:19:57+00:00","author":"Kirat","twitter_card":"summary_large_image","twitter_misc":{"Written by":"Kirat","Est. reading time":"2 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/#article","isPartOf":{"@id":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/"},"author":{"name":"Kirat","@id":"https:\/\/jassweb.com\/solved\/#\/schema\/person\/65c9c7b7958150c0dc8371fa35dd7c31"},"headline":"[Solved] Nvidia Tesla T4 tensor core benchmark [closed]","datePublished":"2022-10-09T00:19:57+00:00","mainEntityOfPage":{"@id":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/"},"wordCount":208,"publisher":{"@id":"https:\/\/jassweb.com\/solved\/#organization"},"keywords":["benchmarking","cuda","gpu","nvidia"],"articleSection":["Solved"],"inLanguage":"en-US"},{"@type":"WebPage","@id":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/","url":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/","name":"[Solved] Nvidia Tesla T4 tensor core benchmark [closed] - JassWeb","isPartOf":{"@id":"https:\/\/jassweb.com\/solved\/#website"},"datePublished":"2022-10-09T00:19:57+00:00","breadcrumb":{"@id":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/jassweb.com\/solved\/solved-nvidia-tesla-t4-tensor-core-benchmark-closed\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/jassweb.com\/solved\/"},{"@type":"ListItem","position":2,"name":"[Solved] Nvidia Tesla T4 tensor core benchmark [closed]"}]},{"@type":"WebSite","@id":"https:\/\/jassweb.com\/solved\/#website","url":"https:\/\/jassweb.com\/solved\/","name":"JassWeb","description":"Build High-quality Websites","publisher":{"@id":"https:\/\/jassweb.com\/solved\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/jassweb.com\/solved\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/jassweb.com\/solved\/#organization","name":"Jass Web","url":"https:\/\/jassweb.com\/solved\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/jassweb.com\/solved\/#\/schema\/logo\/image\/","url":"https:\/\/jassweb.com\/wp-content\/uploads\/2021\/02\/jass-website-logo-1.png","contentUrl":"https:\/\/jassweb.com\/wp-content\/uploads\/2021\/02\/jass-website-logo-1.png","width":693,"height":132,"caption":"Jass Web"},"image":{"@id":"https:\/\/jassweb.com\/solved\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/jassweb.com\/solved\/#\/schema\/person\/65c9c7b7958150c0dc8371fa35dd7c31","name":"Kirat","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/jassweb.com\/solved\/#\/schema\/person\/image\/","url":"https:\/\/jassweb.com\/solved\/wp-content\/litespeed\/avatar\/1261af3c9451399fa1336d28b98ea3bb.jpg?ver=1776403586","contentUrl":"https:\/\/jassweb.com\/solved\/wp-content\/litespeed\/avatar\/1261af3c9451399fa1336d28b98ea3bb.jpg?ver=1776403586","caption":"Kirat"},"sameAs":["http:\/\/jassweb.com"],"url":"https:\/\/jassweb.com\/solved\/author\/jaspritsinghghumangmail-com\/"}]}},"_links":{"self":[{"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/posts\/14806","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/comments?post=14806"}],"version-history":[{"count":0,"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/posts\/14806\/revisions"}],"wp:attachment":[{"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/media?parent=14806"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/categories?post=14806"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/jassweb.com\/solved\/wp-json\/wp\/v2\/tags?post=14806"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}