Research
If AI agents can port GPU kernels in days rather than months, enterprises gain more hardware options and less CUDA lock-in, so infrastructure teams should reassess accelerator sourcing and developers should evaluate AI-assisted kernel generation for non-Nvidia targets.
Anthropic reports that its Claude Mythos model cut the software work needed to onboard AMD GPUs from up to 12 months to around three days. The model reportedly wrote almost all of the GPU kernels autonomously. The result is framed as a challenge to Nvidia's CUDA software moat.
Read the full article at Hacker News
CoFabrix summarises and comments on this story. The original reporting belongs to Hacker News.
Find out where your organization stands -- and what to do about it.