Hardware & Inference RSS

Research 🇺🇸

From CUDA to MLX: How K-Search Brings Decades of Kernel Expertise to Apple Silicon

Researchers from UC Berkeley Sky Lab extended the K-Search evolutionary kernel optimization framework with an MLX backend for Apple Silicon. They developed a structured CUDA-to-MLX translation layer that transfers decades of CUDA optimization expertise to Apple GPUs, achieving near-expert performance on attention and Mamba SSM kernels.

AppleApple NVIDIANVIDIA Google/DeepMindGoogle/DeepMind
BAIR (Berkeley AI)30.07 · 03:03
Fresh news