llama.cpp B9387 Significant AMD/ROCm PP Update
The llama.cpp project has released an update, B9387, which includes significant improvements for AMD ROCm support. This update specifically enables MFMA (Matrix Multiply-Accumulate) operations, but these are currently restricted to AMD's CDNA datacenter cards, including the MI100, MI200, and MI300 series. Users are encouraged to share their initial performance results with this new version. AI
IMPACT Enhances performance for local LLM inference on specific AMD hardware.