Back

Anthropic Says Claude Mythos Cut AMD GPU Onboarding to About Three Days
SiTech AI Team3 min read

Anthropic Says Claude Mythos Cut AMD GPU Onboarding to About Three Days

Anthropic told IPO investors that Claude Mythos Preview did the software work to get its AI running on AMD GPUs in about three days with almost no human intervention, a task it says once took six to 12 months.

Anthropic told IPO investors that its Claude Mythos Preview model handled the software work needed to get its AI running on AMD GPUs in about three days with basically no human intervention. The company said the task previously took six to 12 months and required dozens of engineers. The three-day estimate covers prompt processing for one Anthropic model on AMD MI355X GPUs at a useful level of performance.

Kernel work and Nvidia's CUDA moat

Claude Mythos ported or wrote almost 100 percent of the kernels, the low-level code that ties an AI model to one company's chips. That code has long been part of Nvidia's CUDA moat, and an AI system that can write it in days would suggest that moat is eroding. Tom Brown, Anthropic co-founder and chief compute officer, made the time comparison and noted that the original work used a team of dozens of engineers.

Anthropic's multi-chip strategy

Anthropic already runs its models on GPUs from Nvidia and AMD, Google's TPUs, and AWS Trainium, and manages its fleet as one fungible pool of compute. Brown called the early multi-chip decision an unconventional bet. The company said its historical experience bringing up new accelerators with traditional engineering approaches involved tailoring code to each hardware platform, a challenge that could take six to 12 months.

AMD, Nvidia and remaining limits

Anthropic started working with AMD GPUs for the first time this year and has an agreement for up to 2 gigawatts now in place. AMD's Anush Elangovan told EE Times in April that Claude is better than HIPify because it has web search built in, while AMD CEO Lisa Su said in September that much of AI software written for CUDA can now be translated to AMD in software. Nvidia CEO Jensen Huang said in April that Nvidia uses AI on its own kernels and that nobody knows its architecture better than Nvidia does, and he called Anthropic a unique instance rather than a trend. Anthropic has not said how fast the model ran on AMD GPUs versus Nvidia's, and it has not said whether the replying step was included alongside prompt processing.

What comes next

Anthropic's first AMD MI450 chips in Helios racks are expected in the first half of 2027. Its IPO comes sooner, after which it will report as a public company. The company's prospectus says it can onboard new chip platforms in a fraction of the time it once took, though it gives no number.

Sources: tomshardware.com

SiTech

SiTech — AI-powered web development

We build fast, modern websites and bring AI into real business workflows. Have a project or a question? We'd love to help.