Andrzej Janik announced the release of ZLUDA 6, an open-source implementation of CUDA technology. The project aims to enable unmodified CUDA applications to run on systems with non-NVIDIA GPUs, with performance close to that of applications running without interposers. The project's code is written in Rust and is licensed under the MIT and Apache 2.0 licenses.
ZLUDA 6 became the third significant release of the project, formed after
Cleansing the codebase of code developed during Andrzej's time at AMD. Since 2022, Andrzej had been working at AMD on creating a CUDA compatibility layer for AMD GPUs, but the project was discontinued in 2024. In accordance with the terms of his contract and after receiving permission to publish from an AMD representative, Andrzej open-sourced the work he developed while at AMD, which enables CUDA applications to run on top of the ROCm stack and the Heterogeneous-computing Interface for Portability (HIP) runtime.
Two years ago, Andrzej was forced to remove his code from open access after receiving a letter from lawyers indicating that the permission granted in an email exchange was invalid. Afterward, Andrzej began working on a new version of ZLUDA, based on the codebase that existed before Andrzej joined AMD. Currently, the project focuses on applications that use CUDA to accelerate machine learning tasks. The project is currently limited to AMD GPUs, but will eventually be adapted for Intel GPUs.
The new release adds initial support for running the PhysX physics simulation engine and Blender 3D modeling system on top of ZLUDA. PhysX support is currently in early release status.
Alpha version and limited to running 32-bit PhysX builds, which allows systems with AMD GPUs to run some older games that use this engine. For example, in Mafia II, after enabling ZLUDA, FPS increased from 26 to 80 frames per second. Texture support has been implemented, sufficient for running Blender using ZLUDA. Platform support has been significantly improved. WindowsExpanded capabilities for running frameworks for large language models (such as PyTorch) on top of ZLUDA, leveraging CUDA optimizations for NVIDIA GPUs.
Source: opennet.ru
