{"id":8530,"date":"2018-09-18T08:54:29","date_gmt":"2018-09-18T06:54:29","guid":{"rendered":"http:\/\/blog.netspark.de\/?p=8530"},"modified":"2018-09-14T09:30:33","modified_gmt":"2018-09-14T07:30:33","slug":"nvidia-tesla-m4-p4-and-t4-revealed","status":"publish","type":"post","link":"https:\/\/blog.netspark.de\/?p=8530","title":{"rendered":"NVidia Tesla M4, P4 and T4 revealed!"},"content":{"rendered":"

\"\"It seems as if NVidia takes GPU computing serious. The
\nrecently presented Tesla T4 card, a successor to M4 and
\nP4 (some say they’re graphics cards but that’s only half
\nthe truth) seem to offer tremendous processing power!<\/p>\n

Especially for cryptocoin mining this GPU-based card might be interesting as they offer immense calculation power with only 75W power demand. This means that the cards don’t rely on a separate power connector but can be powered with the PCIe-slot alone.<\/p>\n

NVIDIA Tesla T4 GPU Specifications<\/h3>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Product Name<\/span><\/th>\nTesla M4<\/span><\/th>\nTesla P4<\/span><\/th>\nTesla T4<\/span><\/th>\n<\/tr>\n<\/thead>\n
GPU Architecture<\/strong><\/span><\/td>\nMaxwell GM206<\/span><\/td>\nPascal GP104<\/span><\/td>\nTuring TU104<\/span><\/td>\n<\/tr>\n
GPU Process<\/strong><\/span><\/td>\n28nm<\/span><\/td>\n16nm FinFET<\/span><\/td>\n12nm FinFET<\/span><\/td>\n<\/tr>\n
CUDA Cores<\/strong><\/span><\/td>\n1280 CUDA<\/span><\/td>\n2560 CUDA<\/span><\/td>\n2560 CUDA<\/span><\/td>\n<\/tr>\n
Clock Speed<\/strong><\/span><\/td>\n1072 MHz<\/span><\/td>\n1063 MHz<\/span><\/td>\n1582 MHz<\/span><\/td>\n<\/tr>\n
FP32 Compute<\/strong><\/span><\/td>\n2.20 TFLOPs<\/span><\/td>\n5.50 TFLOPs<\/span><\/td>\n8.1 TFLOPs<\/span><\/td>\n<\/tr>\n
FP16 Compute<\/strong><\/span><\/td>\nN\/A<\/span><\/td>\n11 TFLOPs<\/span><\/td>\n65 TFLOPs<\/span><\/td>\n<\/tr>\n
INT8 Compute<\/strong><\/span><\/td>\nN\/A<\/span><\/td>\n22 DLTOPs<\/span><\/td>\n130 DLTOPs<\/span><\/td>\n<\/tr>\n
INT4 Compute<\/strong><\/span><\/td>\nN\/A<\/span><\/td>\nN\/A<\/span><\/td>\n260 DLTOPs<\/span><\/td>\n<\/tr>\n
VRAM<\/strong><\/span><\/td>\n4 GB GDDR5<\/span><\/td>\n8 GB GDDR5<\/span><\/td>\n16 GB GDDR6<\/span><\/td>\n<\/tr>\n
Memory Clock<\/strong><\/span><\/td>\n5.5 GHz<\/span><\/td>\n6.0 GHz<\/span><\/td>\n10 GHz<\/span><\/td>\n<\/tr>\n
Memory Bus<\/strong><\/span><\/td>\n128-bit<\/span><\/td>\n256-bit<\/span><\/td>\n256-bit<\/span><\/td>\n<\/tr>\n
Memory Bandwidth<\/strong><\/span><\/td>\n88.0 GB\/s<\/span><\/td>\n192.0 GB\/s<\/span><\/td>\n320 GB\/s+<\/span><\/td>\n<\/tr>\n
TDP<\/strong><\/span><\/td>\n~75W<\/span><\/td>\n~75W<\/span><\/td>\n75W<\/span><\/td>\n<\/tr>\n
Launch<\/strong><\/span><\/td>\n2015<\/span><\/td>\n2016<\/span><\/td>\nQ4 2018<\/span><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

Impressive. However there’s no word spoken about the price, the slides may promise a lot of bang for the buck and who knows… maybe it’ll bring back “classic” GPU cards into more humane price regions as the current RTX models start at 900 US$ and end up to 1800US$. And the 1800 US$ model isn’t a RTX Titan or whatever the card would be named. It’s just a RTX 2080Ti…<\/p>\n

The Turing based NVIDIA Tesla T4 graphics card is aimed at inference acceleration markets. It is designed to give deep learning performance a massive boost over its predecessors and is also going to deliver immense performance for AI video applications. NVIDIA\u2019s own estimate put the graphics card at twice as fast in video processing, allowing users to decode up to 38(!) full-HD video streams at the same time which just wasn\u2019t possible on the previous generation.<\/p>\n

The NVIDIA Tesla T4 GPU is the world\u2019s most advanced inference accelerator. Powered by NVIDIA Turing Tensor Cores, T4 brings revolutionary multi-precision inference performance to accelerate the diverse applications of modern AI. Packaged in an energy-efficient 75-watt, small PCIe form factor, T4 is optimized for scale-out servers and is purpose-built to deliver state-of-the-art inference in real time.<\/p>\n

As the volume of online videos continues to grow exponentially, demand for solutions to efficiently search and gain insights from video continues to grow as well. Tesla T4 delivers breakthrough performance for AI video applications, with dedicated hardware transcoding engines that bring twice the decoding performance of prior-generation GPUs. T4 can decode up to 38 full-HD video streams, making it easy to integrate scalable deep learning into video pipelines to deliver innovative, smart video services.<\/p>\n

Jensen Huang, CEO of NVIDIA<\/p>\n<\/div>\n

\"\"<\/a>Jensen Huang, at the GTC 2018 Japan keynote presenting the Tesla T4<\/p>\n

\"\"<\/a>Product shot of the Tesla T4<\/p>\n

\"\"<\/a>Performance chart<\/p>\n

\"\"<\/a>Turing Tensor Cores scheme<\/p>\n

There’s no price tag available yet nor when it’ll become in reasonable numbers. But if they get available by then, maybe we will see a sligt change in pricing for the RTX GPU series. I would greatly anticipate a price slide downwards!<\/p>\n","protected":false},"excerpt":{"rendered":"

It seems as if NVidia takes GPU computing serious. The recently presented Tesla T4 card, a successor to M4 and P4 (some say they’re graphics cards but that’s only half the truth) seem to offer tremendous processing power!<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[55,3,2949],"tags":[3181,2975,3180,347,3179,3178,3182],"class_list":["post-8530","post","type-post","status-publish","format-standard","hentry","category-computer-2","category-news","category-technology","tag-ai-acceleration","tag-cryptomining","tag-deep-learning","tag-nvidia","tag-tensor-core","tag-tesla-t4","tag-turing"],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/blog.netspark.de\/index.php?rest_route=\/wp\/v2\/posts\/8530","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blog.netspark.de\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.netspark.de\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.netspark.de\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.netspark.de\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=8530"}],"version-history":[{"count":0,"href":"https:\/\/blog.netspark.de\/index.php?rest_route=\/wp\/v2\/posts\/8530\/revisions"}],"wp:attachment":[{"href":"https:\/\/blog.netspark.de\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8530"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.netspark.de\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8530"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.netspark.de\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8530"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}