--- title: AMD’s Helios puts 72 GPUs and 31 terabytes of HBM4 in... description: AMD has unveiled Helios, a rack-scale AI system packing 72 Instinct MI455X GPUs, 31TB of HBM4 memory, and 2.9 exaflops of FP4 inference compute into a single... canonical: https://daily.dev/posts/amd-s-helios-puts-72-gpus-and-31-terabytes-of-hbm4-in-one-rack-it-is-amd-s-answer-to-nvidia-s-nvl72-rt5ma3xnm twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AMD’s Helios puts 72 GPUs and 31 terabytes of HBM4 in one rack. It is AMD’s answer to Nvidia’s NVL72. | daily.dev og:description: AMD has unveiled Helios, a rack-scale AI system packing 72 Instinct MI455X GPUs, 31TB of HBM4 memory, and 2.9 exaflops of FP4 inference compute into a single... og:url: https://daily.dev/posts/amd-s-helios-puts-72-gpus-and-31-terabytes-of-hbm4-in-one-rack-it-is-amd-s-answer-to-nvidia-s-nvl72-rt5ma3xnm og:image: https://api.daily.dev/og/posts/RT5MA3xnM.png og:image:alt: AMD’s Helios puts 72 GPUs and 31 terabytes of HBM4 in one rack. It is AMD’s answer to Nvidia’s NVL72. og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # AMD’s Helios puts 72 GPUs and 31 terabytes of HBM4 in one rack. It is AMD’s answer to Nvidia’s NVL72. **[The Next Web](https://daily.dev/sources/tnw)** · 3 min read · 1 upvotes · 0 comments ## Summary AMD has unveiled Helios, a rack-scale AI system packing 72 Instinct MI455X GPUs, 31TB of HBM4 memory, and 2.9 exaflops of FP4 inference compute into a single rack. Built on 18 compute trays each combining four MI455X accelerators (CDNA 5 architecture) with a sixth-gen EPYC Venice CPU, Helios directly challenges Nvidia's Vera Rubin NVL72. A key differentiator is AMD's bet on open standards — UALink for GPU interconnect, Ultra Ethernet Consortium specs for scale-out networking, and OCP Open Rack Wide form factor — versus Nvidia's proprietary NVLink. The full rack delivers 260 TB/s of scale-up bandwidth and 43 TB/s of scale-out bandwidth. Engineering samples are expected in H2 2026, with mass production in Q2 2027. The ROCm software stack supports PyTorch, TensorFlow, and JAX, though closing Nvidia's CUDA ecosystem advantage remains the critical challenge. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How many GPUs and how much memory does AMD's Helios rack system have? AMD's Helios packs 72 Instinct MI455X GPUs and 31 terabytes of HBM4 memory into a single rack, delivering 2.9 exaflops of FP4 inference compute and 1.4 exaflops of FP8 training. Each rack uses 18 compute trays holding four MI455X GPUs and one sixth-generation EPYC 'Venice' CPU. Engineering samples ship in the second half of 2026, with mass production starting Q2 2027. _Teams weighing AMD against Nvidia for AI infrastructure can follow spec comparisons like this on daily.dev._ ### What interconnect standard does AMD Helios use instead of Nvidia's NVLink? AMD Helios uses UALink for scale-up interconnect between GPUs within the rack and Ultra Ethernet Consortium specifications for scale-out networking between racks, along with the OCP Open Rack Wide form factor. This contrasts with Nvidia's NVL72, which relies on the proprietary NVLink interconnect, giving data centre operators an open-standards alternative to avoid vendor lock-in. _Anyone comparing open versus proprietary AI interconnect standards can track this shift on daily.dev._ ### Does AMD's ROCm software stack support PyTorch and other frameworks used on Nvidia GPUs? Yes, AMD's ROCm software stack supports PyTorch, TensorFlow, and JAX, meaning developers do not need to rewrite code to move workloads from Nvidia's CUDA ecosystem, at least in theory. Whether AMD can fully close the software gap that has historically kept it behind Nvidia in AI compute remains the open question that its new Helios hardware is designed to test. _Developers evaluating a CUDA-to-ROCm migration can keep up with ecosystem maturity news on daily.dev._ ## Similar posts on daily.dev - [AMD taking AI fight to Nvidia with Helios rack-scale system](https://daily.dev/posts/amd-taking-ai-fight-to-nvidia-with-helios-rack-scale-system-5npotliep) · The Register · 0 upvotes · 0 comments - [AMD is investing $5 billion in Anthropic and deploying 2 gigawatts of Helios GPUs to run Claude](https://daily.dev/posts/amd-is-investing-5-billion-in-anthropic-and-deploying-2-gigawatts-of-helios-gpus-to-run-claude-ucsqlzw5e) · The Next Web · 0 upvotes · 0 comments --- Tags: [#nvidia](https://daily.dev/tags/nvidia), [#gpu](https://daily.dev/tags/gpu), [#amd](https://daily.dev/tags/amd), [#ai-infrastructure](https://daily.dev/tags/ai-infrastructure) [View this post on daily.dev](https://daily.dev/posts/amd-s-helios-puts-72-gpus-and-31-terabytes-of-hbm4-in-one-rack-it-is-amd-s-answer-to-nvidia-s-nvl72-rt5ma3xnm) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"AMD’s Helios puts 72 GPUs and 31 terabytes of HBM4 in one rack. It is AMD’s answer to Nvidia’s NVL72.","url":"https://daily.dev/posts/amd-s-helios-puts-72-gpus-and-31-terabytes-of-hbm4-in-one-rack-it-is-amd-s-answer-to-nvidia-s-nvl72-rt5ma3xnm","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/amd-s-helios-puts-72-gpus-and-31-terabytes-of-hbm4-in-one-rack-it-is-amd-s-answer-to-nvidia-s-nvl72-rt5ma3xnm"},"datePublished":"2026-07-20T15:31:51.714Z","dateModified":"2026-09-14T06:39:13.296Z","description":"AMD has unveiled Helios, a rack-scale AI system packing 72 Instinct MI455X GPUs, 31TB of HBM4 memory, and 2.9 exaflops of FP4 inference compute into a single...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/3c70621e8f22a7ffad2882e61f96837f?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/3c70621e8f22a7ffad2882e61f96837f?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The Next Web","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The Next Web","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/tnw","url":"https://daily.dev/sources/tnw"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/amd-s-helios-puts-72-gpus-and-31-terabytes-of-hbm4-in-one-rack-it-is-amd-s-answer-to-nvidia-s-nvl72-rt5ma3xnm","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"nvidia,gpu,amd,ai-infrastructure","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The Next Web","item":"https://daily.dev/sources/tnw"},{"@type":"ListItem","position":3,"name":"AMD’s Helios puts 72 GPUs and 31 terabytes of HBM4 in one rack. It is AMD’s answer to Nvidia’s NVL72."}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/amd-s-helios-puts-72-gpus-and-31-terabytes-of-hbm4-in-one-rack-it-is-amd-s-answer-to-nvidia-s-nvl72-rt5ma3xnm#faq","mainEntity":[{"@type":"Question","name":"How many GPUs and how much memory does AMD's Helios rack system have?","acceptedAnswer":{"@type":"Answer","text":"AMD's Helios packs 72 Instinct MI455X GPUs and 31 terabytes of HBM4 memory into a single rack, delivering 2.9 exaflops of FP4 inference compute and 1.4 exaflops of FP8 training. Each rack uses 18 compute trays holding four MI455X GPUs and one sixth-generation EPYC 'Venice' CPU. Engineering samples ship in the second half of 2026, with mass production starting Q2 2027. Teams weighing AMD against Nvidia for AI infrastructure can follow spec comparisons like this on daily.dev."}},{"@type":"Question","name":"What interconnect standard does AMD Helios use instead of Nvidia's NVLink?","acceptedAnswer":{"@type":"Answer","text":"AMD Helios uses UALink for scale-up interconnect between GPUs within the rack and Ultra Ethernet Consortium specifications for scale-out networking between racks, along with the OCP Open Rack Wide form factor. This contrasts with Nvidia's NVL72, which relies on the proprietary NVLink interconnect, giving data centre operators an open-standards alternative to avoid vendor lock-in. Anyone comparing open versus proprietary AI interconnect standards can track this shift on daily.dev."}},{"@type":"Question","name":"Does AMD's ROCm software stack support PyTorch and other frameworks used on Nvidia GPUs?","acceptedAnswer":{"@type":"Answer","text":"Yes, AMD's ROCm software stack supports PyTorch, TensorFlow, and JAX, meaning developers do not need to rewrite code to move workloads from Nvidia's CUDA ecosystem, at least in theory. Whether AMD can fully close the software gap that has historically kept it behind Nvidia in AI compute remains the open question that its new Helios hardware is designed to test. Developers evaluating a CUDA-to-ROCm migration can keep up with ecosystem maturity news on daily.dev."}}]} ```