ZLUDA, an open-source project enabling CUDA code execution on non-NVIDIA GPUs, has expanded to two full-time developers and made significant progress in Q2 2025. The project now supports multi-vendor GPU implementations for AI workloads, handles ABI breaks in ROCm, ensures bit-accurate execution across different hardware, and has begun supporting llm.c for large language model training. ZLUDA also added automated GitHub builds and improved logging capabilities.

1m read timeFrom phoronix.com
Post cover image

Sort: