![\"\"](\"https:\/\/blog.netspark.de\/wp-content\/uploads\/2018\/08\/AMDRadeon.png\")
HPC (High Performance Computing) is a highly demanded\nthing today. When it comes to pure computing power for
\nvast numbercrunching, then classic CPUs cannot achieve
\nthe speed modern GPUs can reach today.<\/p>\n
HPC (High Performance Computing) is a highly demanded
\nthing today. When it comes to pure computing power for
\nvast numbercrunching, then classic CPUs cannot achieve
\nthe speed modern GPUs can reach today.<\/p>\n
AMD just released the infor that the Instinct MI50 and MI60 are on their way as successor to the Vega MI25.<\/p>\n
Fabbed at 7nm they squeeze 13 billion transistors onto a diesize of only 334mm2<\/sup>, which gives room for more speculation. As an example the Titan V from NVidia is still fabbed at 10nm and thus their HPC monster has a GPU chip with over 600mm2<\/sup> offering “only” 11 billion transistors.<\/p>\n But enough babbling about die sizes. What does that mean in numbers?<\/p>\n Well, the computing power is measured in FLOPS (Floating Point Operations per seconds). With this chip series, also Integer 8 Byte (INT8) und Integer 4 Byte (INT4) instructions can be processed blazing fast.<\/p>\n In raw numbers:<\/p>\n FP64:\u00a0\u00a0\u00a0\u00a0 7.4 TFLOPS Although the FP16-Arithmetics didn’t get a high boost (roughly 20%) the usage of 32GB HBM2 memory allowed the new Instinct cards to perform up to 280% percent faster in benchmarks like Resnet-50, Inception-4 and VGG16 compared to the MI25 cards.<\/p>\n So whatever comes, maybe also Bitcoin and Ethereum miners will sort of benefit from the new computing cards. The release is scheduled for Q4 2018(!) but prices aren’t published yet!<\/p>\n Oh, and one more thing: You definitely will need a PICe4.0 compatible board and processor for these beasts. So you’d better watch for the upcoming EPYC Rome processors from AMD which will natively support PCIe4.0<\/p>\n","protected":false},"excerpt":{"rendered":" HPC (High Performance Computing) is a highly demanded thing today. When it comes to pure computing power for vast numbercrunching, then classic CPUs cannot achieve the speed modern GPUs can reach today.<\/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":[177,2158,3354,3353,3352,3355,1689,3351],"class_list":["post-19023","post","type-post","status-publish","format-standard","hentry","category-computer-2","category-news","category-technology","tag-amd","tag-computing","tag-flops","tag-hpc","tag-instinct","tag-iops","tag-performance","tag-radeion"],"jetpack_featured_media_url":"","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/blog.netspark.de\/index.php?rest_route=\/wp\/v2\/posts\/19023","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=19023"}],"version-history":[{"count":0,"href":"https:\/\/blog.netspark.de\/index.php?rest_route=\/wp\/v2\/posts\/19023\/revisions"}],"wp:attachment":[{"href":"https:\/\/blog.netspark.de\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=19023"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.netspark.de\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=19023"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.netspark.de\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=19023"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\nFP32:\u00a0\u00a0 14.7 TFLOPS
\nFP16:\u00a0\u00a0 29.5 TFLOPS
\nINT8:\u00a0\u00a0 59.2 IOPS
\nINT4: 118.0 IOPS<\/p>\n