From CUDA to MLX: K-Search Brings Decades of Kernel Expertise to Apple Silicon
Berkeley researchers have extended the K-Search evolutionary kernel search framework with an MLX backend, enabling the adaptation of CUDA kernel optimizations to Apple Silicon. Their approach achieves near-expert performance, with 0.97x speedup compared to the native MLX Attention kernel and up to 20x prefill speedup over baseline implementations.
Why it matters: This work could significantly reduce the engineering effort required to port high-performance GPU kernels across hardware platforms, accelerating AI deployment on non-NVIDIA devices.
Full story at: Berkeley AI Research ↗