What changed
Anthropic announced that it has developed a multi-chip strategy allowing it to train and run its Claude model on all four major chip families, including AMD GPUs. The company claims that its frontier model, Mythos, ported or wrote nearly 100 percent of the kernels for AMD GPUs in just three days, instead of the usual six to twelve months.
What this means in practice
Anthropic's approach involves treating compute from different vendors as a single, interchangeable pool. This allows the company to allocate computing capacity across AWS Trainium, Google TPUs, Nvidia GPUs, and AMD GPUs for tasks such as research, model development, and internal Claude usage. According to Tom Brown, co-founder and Chief Compute Officer at Anthropic, the company has become "incredibly good" at this multi-chip strategy after years of investment.
Limitations and availability
The figures provided by Anthropic refer to prompt processing of an Anthropic model on AMD MI355X GPUs. The company claims that its AI model, Mythos, brought Claude to a usable performance level on AMD hardware in three days with almost no human intervention. However, the supplied announcement does not specify whether this process is fully automated or if further human oversight is required. The figures come from Anthropic's investor presentation and have not been independently verified.
