AMD Ryzen AI NPUs Are Finally Useful Under Linux For Running LLMs

Phoronix
Lemonade 10.0, utilizing the FastFlowLM runtime, now enables AMD Ryzen AI NPUs to effectively run Large Language Models on Linux.

Summary

For two years, support for AMD Ryzen AI NPUs in the Linux kernel via the AMDXDNA driver has existed, but user-space software utilization has been severely limited. This has significantly changed with the release of Lemonade 10.0, which introduces Linux NPU support for running Large Language Models (LLMs) and Whisper, alongside native integration with Claude Code. Lemonade leverages the open-source FastFlowLM runtime, which was built specifically as an NPU-first solution for Ryzen AI, capable of handling context lengths up to 256k tokens with current-gen NPUs. To use this feature, users require the Linux 7.0 kernel or back-ported AMDXDNA driver updates, and it supports all current AMD Ryzen AI 300/400 series SoCs. This development is timely, especially as the Ryzen AI Embedded P100 and PRO 400 series are entering markets where Linux adoption is expected to be higher.

(Source:Phoronix)