{"id":14407,"date":"2024-06-03T11:26:55","date_gmt":"2024-06-03T10:26:55","guid":{"rendered":"https:\/\/roglacup.com\/klaus62\/?p=14407"},"modified":"2024-06-03T11:58:55","modified_gmt":"2024-06-03T10:58:55","slug":"amd-targets-nvidia-with-288-gb-instinct-mi325x-gpu-coming-this-year","status":"publish","type":"post","link":"https:\/\/roglacup.com\/klaus62\/2024\/06\/03\/amd-targets-nvidia-with-288-gb-instinct-mi325x-gpu-coming-this-year\/","title":{"rendered":"AMD Targets Nvidia With 288-GB Instinct MI325X GPU Coming This Year"},"content":{"rendered":"\n<p><a href=\"https:\/\/twitter.com\/intent\/tweet?text=AMD%20reveals%20the%20MI325X%2c%20a%20288GB%20AI%20accelerator%20built%20to%20battle%20Nvidia%27s%20H200&amp;url=https:\/\/www.theregister.com\/2024\/06\/03\/amd_reveals_refreshed_mi325x_with\/%3futm_medium%3dshare%26utm_content%3darticle%26utm_source%3dtwitter&amp;via=theregister\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><a href=\"https:\/\/www.facebook.com\/dialog\/feed?app_id=1404095453459035&amp;display=popup&amp;link=https:\/\/www.theregister.com\/2024\/06\/03\/amd_reveals_refreshed_mi325x_with\/%3futm_medium%3dshare%26utm_content%3darticle%26utm_source%3dfacebook\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><a href=\"https:\/\/www.linkedin.com\/shareArticle?mini=true&amp;url=https:\/\/www.theregister.com\/2024\/06\/03\/amd_reveals_refreshed_mi325x_with\/%3futm_medium%3dshare%26utm_content%3darticle%26utm_source%3dlinkedin&amp;title=AMD%20reveals%20the%20MI325X%2c%20a%20288GB%20AI%20accelerator%20built%20to%20battle%20Nvidia%27s%20H200&amp;summary=House%20of%20Zen%20hopes%20to%20catch%20its%20AI%20rival%20by%20moving%20to%20a%20yearly%20release%20cadence\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><a href=\"https:\/\/api.whatsapp.com\/send?text=https:\/\/www.theregister.com\/2024\/06\/03\/amd_reveals_refreshed_mi325x_with\/%3futm_medium%3dshare%26utm_content%3darticle%26utm_source%3dwhatsapp\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><\/p>\n\n\n\n<p><strong>COMPUTEX<\/strong>&nbsp;AMD&#8217;s flagship AI accelerator will receive a high-bandwidth boost when its MI325X arrives later this year.<\/p>\n\n\n\n<p>The reveal comes as AMD follows Nvidia&#8217;s pattern and transitions to a yearly release cadence for its &#8220;Instinct&#8221; range of accelerators.<\/p>\n\n\n\n<p>The Instinct MI325X, at least from what we can tell, is a lot like Nvidia&#8217;s H200 in that it&#8217;s a HBM3e-enhanced version of the GPU we&nbsp;<a target=\"_blank\" href=\"https:\/\/www.theregister.com\/2023\/12\/06\/amd_mi300_gpu\/\" rel=\"noreferrer noopener\">detailed<\/a>&nbsp;at length during AMD&#8217;s Advancing AI event in December 2023. But the part is among the most sophisticated we&#8217;ve seen to date \u2013 composed of eight compute, four I\/O, and eight memory chiplets stitched together using a combination of 2.5D and 3D packaging technologies.<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"509\" src=\"https:\/\/roglacup.com\/klaus62\/wp-content\/uploads\/2024\/06\/image-1024x509.png\" alt=\"\" class=\"wp-image-14410\" srcset=\"https:\/\/roglacup.com\/klaus62\/wp-content\/uploads\/2024\/06\/image-1024x509.png 1024w, https:\/\/roglacup.com\/klaus62\/wp-content\/uploads\/2024\/06\/image-300x149.png 300w, https:\/\/roglacup.com\/klaus62\/wp-content\/uploads\/2024\/06\/image-768x382.png 768w, https:\/\/roglacup.com\/klaus62\/wp-content\/uploads\/2024\/06\/image.png 1200w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><a href=\"https:\/\/regmedia.co.uk\/2024\/06\/02\/supplied_amd_mi325x_half_lidded.jpg\" target=\"_blank\" rel=\"noreferrer noopener\"><\/a><\/p>\n\n\n\n<p>AMD&#8217;s MI325X accelerator half lidded<\/p>\n\n\n\n<p>From what we&#8217;ve seen, it doesn&#8217;t appear that the CDNA 3 GPU tiles powering the forthcoming chip have changed meaningfully \u2013 at least not in terms of FLOPS. The chip still boasts 1.3 petaFLOPS of dense BF\/FP16 performance or 2.6 petaFLOPS when dropping down to FP8. To be clear, the MI325X is still faster than the H200 at any given precision.<\/p>\n\n\n\n<p>AMD&#8217;s focus seems to be extending its memory advantage over Nvidia. At launch, the 192GB MI300X boasted more than twice the HBM3 of the H100 and a 51GB edge over the upcoming H200. The MI325X boosts the accelerator&#8217;s capacity to 288GB \u2013 more than twice that of the H200 and 50 percent more than Nvidia&#8217;s Blackwell chips revealed at GTC this spring.<\/p>\n\n\n\n<p>The move to HBM3e also juices the MI325X&#8217;s memory bandwidth to 6TB\/sec. While a decent boost from the 5.3TB\/sec of the MI300X and 1.3x more than the H200, we would have expected that number to be closer to 8TB\/sec \u2013 like&nbsp;<a target=\"_blank\" href=\"https:\/\/www.theregister.com\/2024\/03\/18\/nvidia_turns_up_the_ai\/?td=rt-9c\" rel=\"noreferrer noopener\">we saw<\/a>&nbsp;on Nvidia&#8217;s Blackwell GPUs.<\/p>\n\n\n\n<p>Unfortunately, we&#8217;ll have to wait until the MI325X arrives later this year to find out what&#8217;s going on with its memory config.<\/p>\n\n\n\n<p>Posted from: <a href=\"https:\/\/www.theregister.com\/2024\/06\/03\/amd_reveals_refreshed_mi325x_with\/\">https:\/\/www.theregister.com\/2024\/06\/03\/amd_reveals_refreshed_mi325x_with\/<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>COMPUTEX&nbsp;AMD&#8217;s flagship AI accelerator will receive a high-bandwidth boost when its MI325X arrives later this year. The reveal comes as AMD follows Nvidia&#8217;s pattern and transitions to a yearly release cadence for its &#8220;Instinct&#8221; range of accelerators. The Instinct MI325X, at least from what we can tell, is a lot like Nvidia&#8217;s H200 in that&hellip;&nbsp;<a href=\"https:\/\/roglacup.com\/klaus62\/2024\/06\/03\/amd-targets-nvidia-with-288-gb-instinct-mi325x-gpu-coming-this-year\/\" class=\"\" rel=\"bookmark\">Read More &raquo;<span class=\"screen-reader-text\">AMD Targets Nvidia With 288-GB Instinct MI325X GPU Coming This Year<\/span><\/a><\/p>\n","protected":false},"author":1,"featured_media":14410,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"neve_meta_sidebar":"","neve_meta_container":"","neve_meta_enable_content_width":"","neve_meta_content_width":0,"neve_meta_title_alignment":"","neve_meta_author_avatar":"","neve_post_elements_order":"","neve_meta_disable_header":"","neve_meta_disable_footer":"","neve_meta_disable_title":"","footnotes":""},"categories":[812,1156],"tags":[],"class_list":["post-14407","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-amd"],"amp_enabled":true,"_links":{"self":[{"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/posts\/14407","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/comments?post=14407"}],"version-history":[{"count":2,"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/posts\/14407\/revisions"}],"predecessor-version":[{"id":14411,"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/posts\/14407\/revisions\/14411"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/media\/14410"}],"wp:attachment":[{"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/media?parent=14407"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/categories?post=14407"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/roglacup.com\/klaus62\/wp-json\/wp\/v2\/tags?post=14407"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}