Dispatch
CUDA-for-AMD-Windows chains ZLUDA and ROCm to run CUDA workloads on a Radeon RX 9060 XT
The project wires CUDA application to ZLUDA to a cuBLAS/cuSPARSE/cuFFT shim to rocBLAS/hipBLASLt/rocSPARSE/HIP, using ZLUDA v6-preview.69, AMD HIP SDK 6.4 and LibTorch 2.3.0 against CUDA 11.8, and validates it by training a 2.2M-parameter PPO network. Only the RX 9060 XT (gfx1200) is verified; cuDNN is unavailable in the stable Windows HIP SDK, and NCCL, TensorRT and custom CUDA extensions can fail. At 110 stars and nine commits it is early, but it is the most concrete published recipe for CUDA-on-AMD under Windows rather than Linux.
↳ Follow the thread