Anthropic says Claude Mythos Preview onboarded AMD MI355X GPUs in about three days, versus six to 12 months
Anthropic told IPO investors that Claude Mythos Preview did the software work to run its AI on AMD MI355X GPUs in about three days, a job it says historically took six to 12 months. The three days cover prompt processing for one model only, and no performance comparison with Nvidia has been given.
Anthropic says Claude Mythos Preview did the software work to get its AI running on AMD's MI355X GPUs in about three days, with "basically no human intervention." The company says the job used to take six to 12 months. The claim comes from Anthropic's IPO roadshow video, dated Nov. 10, as reported by Tom's Hardware.
What Anthropic claims
Tom Brown, Anthropic's co-founder and chief compute officer, made the comparison to investors. He said the original effort meant a team of "dozens of engineers." According to the roadshow video, Claude "ported or wrote almost 100% of the kernels."
A slide footnote narrows the scope. The three days cover "prompt processing for one Anthropic model on AMD MI355X GPUs at a useful level of performance." The six-to-12-month figure is an estimate drawn from Anthropic's "historical experience bringing up new accelerators using traditional engineering approaches."
Anthropic's prospectus (S-1/A No. 2, Nov. 10) makes a related statement without a number: "We can onboard new chip platforms in a fraction of the time it once took." It does not name a chip.
Background
Anthropic runs its models on Nvidia and AMD GPUs, Google's TPUs and AWS Trainium, and manages them as "one fungible pool of compute." Brown called the early multi-chip decision "an unconventional bet." Anthropic began working with AMD GPUs this year and has an agreement with AMD for up to 2 gigawatts.
Kernels are the low-level code that ties a model to a specific vendor's chips. Nvidia's CUDA platform is nearly two decades old. AMD's equivalent is ROCm, with HIP as its programming layer, and AMD built a conversion tool called HIPify. Tom's Hardware notes that a conversion cost to switching still exists.
The Nov. 10 claim follows earlier statements from AMD's side:
- April 1, 2026: AMD VP of AI software Anush Elangovan told EE Times, "Claude is better than HIPify because it has web search built in."
- July 23, 2026, AMD Advancing AI: Brown said one engineer asked Claude to bring up an MI355X rack and left it running "over the weekend." AMD SVP of AI Vamsi Boppana said that in some cases AI-generated kernels were "better than the most manually tuned versions."
- Same event: OpenAI's Philippe Tillet said "agents are taking over our jobs, in a sense, as kernel engineers."
- April 15, 2026: Nvidia CEO Jensen Huang told Dwarkesh Patel, "We use a ton of AI to create the kernels that we have," and that "nobody knows our architecture better than we do."
Under its July agreement with Anthropic, AMD is also using Claude to optimize workloads for its Instinct GPUs and to speed up ROCm development.
What is still unclear
Anthropic has not said whether the decode step, where a model generates its reply, was included alongside prompt processing (prefill). Its chart, with the axis labeled "Time to onboard new silicon," shows both lines reaching a "~100%" mark without saying 100% of what. Anthropic also has not said how fast the model ran on AMD's GPUs compared with Nvidia's.
What this means
This is the first time the onboarding claim has come from Anthropic to its own investors rather than from AMD or its stage. Anthropic buys chips rather than selling them, which Tom's Hardware says gives its numbers more weight than a chipmaker's.
The claim is still narrow. It covers one model, one workload phase and a performance level described only as "useful." Without decode coverage and a speed comparison against Nvidia, the figures do not show whether AMD hardware running AI-written kernels can match Nvidia silicon. Those are the comparisons Tom's Hardware says it wants to see. The six-to-12-month baseline is Anthropic's own historical estimate, not an independent measurement.
Related Articles
Anthropic claims Mythos ported Claude to AMD GPUs in three days, versus six to twelve months
Anthropic claims in an investor presentation that its frontier model Mythos wrote or ported nearly 100 percent of the kernels needed to run Claude on AMD GPUs, taking three days instead of the usual six to twelve months. The company says it trains and runs Claude on four major chip families. The figures have not been independently verified.
VulnCheck: Project Glasswing has 37,966 candidate vulnerabilities, 217 marked fixed in public records
Security firm VulnCheck reports that seven months after launch, Anthropic's Project Glasswing has produced 37,966 candidate vulnerabilities, with 217 marked fixed in public disclosure records. Anthropic's own figures, as of November 2, show 629 upstream patches, and both sides point to human review as the bottleneck.
Claude can now build HTML-in-Canvas sites, needs Chrome or Edge 157 or Claude desktop 5.11492
Claude can now build web pages that embed real HTML elements such as text and buttons inside canvas graphics and 3D effects. The feature works in Chrome and Edge from version 157 and in Claude's desktop apps from version 5.11492. Users must request HTML-in-Canvas explicitly in their prompt.
Anthropic: Selective notes beat full logs for Claude agent memory, with 'Dreaming' fixing errors
Anthropic published research on how Claude retains knowledge across sessions using note files. Its tests show selective notes outperform exhaustive archives, and a post-session process called Dreaming can correct faulty notes. All results come from Anthropic's own tests.
Comments
Loading...