Solo dev enables running CUDA on AMD hardware in Windows, getting multiple CUDA libraries running on a gaming Radeon RX 9060 XT GPU in Windows — CUDA-exclusive workloads on AMD hardware in Windows possible without virtualization or dual-booting
A developer has wired up the ZLUDA project to AMD's HIP libraries for Windows, putting a small but signifcant bridge over NVIDIA's CUDA moat.
AMD has recently added official PyTorch and HIP SDK support for Windows, covering the Radeon RX 7000 and RX 9000 series. However, running proprietary applications, older repositories, or specialized AI tools that only support NVIDIA CUDA on AMD hardware in Windows has been challenging. A new project, Speedstu's CUDA-for-AMD-Windows, could bridge this gap.
The project is a PowerShell setup that connects ZLUDA, a translation layer, with AMD's native HIP/ROCm SDK for Windows. It automatically detects the GPU architecture, grabs a specific version of ZLUDA, and maps it to the ROCm math libraries. The toolkit successfully maps several CUDA libraries, including cuBLAS, cuSPARSE, and cuFFT, to their AMD equivalents.
While it can run CUDA workloads on AMD hardware, cuDNN is not yet supported. The project is a solo open-source effort, meaning its capabilities are limited and depend on the specific workload. Despite this, the project proves that running CUDA-exclusive software on AMD GPUs isn't an insurmountable challenge, but rather a tooling issue that can be addressed through translation.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
