AMD has introduced ROCm.AI, a platform intended to make it easier to optimise AI models and inference workloads for its Instinct GPUs using AI coding assistants. The move addresses the long-standing perception that Nvidia’s CUDA software ecosystem makes its hardware easier to use and faster in practice, even where portable frameworks such as PyTorch and JAX can run on both vendors’ chips.

ROCm.AI supplies coding assistants with AMD-specific tools and documentation for deployment, debugging and performance tuning, while its Hyperloom system can benchmark workloads, find bottlenecks and alter configurations or generate kernels. AMD claims Hyperloom improved performance by 38 per cent over baseline tests on its new Helios racks. The platform will be available through a command-line interface and plug-ins for Claude Code, OpenAI Codex, Google Antigravity and Cursor.