ZLUDA 6 runs unmodified CUDA on AMD GPUs

Category: Dev tools
Status: confirmed
Sources: ZLUDA release notes, discussion
Summary: ZLUDA, the compatibility layer that runs unmodified CUDA applications on AMD GPUs through ROCm, published Version 6 on 2026-06-29 (equivalent to preview build 6-preview.79). The release adds pre-alpha 32-bit PhysX support so older games using PhysX effects run on AMD hardware, basic texture support that makes Blender functional, and incremental machine-learning improvements through new GPU instructions, compiler bug fixes, and cuBLAS and cuDNN work for PyTorch workloads. The Windows loader now auto-loads performance libraries and prints explicit installation instructions when they are missing. The author lists limitations: PhysX fluid simulation remains glitchy and requires source compilation, texture support is minimal, and Windows users must still source ROCm dependencies from the outdated official SDK or buggy nightly builds.
Comments: HN commenters noted the project previously took AMD funding before a change of direction, asked how it compares to Vulkan backends for LLM inference, and observed that several Version 6 features had earlier been declared out of scope.
Why it matters: A maintained path to run CUDA binaries on non-NVIDIA GPUs lowers vendor lock-in for compute and games, part of the broader pressure to break the CUDA moat alongside non-NVIDIA training efforts.