Andrzej Janik presented the release of ZLUDA 6, an open implementation of the CUDA. The goal of the project is to enable unmodified CUDA applications to run on systems with non-NVIDIA GPUs with performance close to that of bare-bones applications. The project code is written in Rust and distributed under the MIT and Apache 2.0 licenses.
ZLUDA 6 was the third significant release of the project, formed after cleaning the code base from code developed during Andrzej’s time at the company AMD. Since 2022, Andrzej has been working at AMD to create a layer for compatibility of AMD GPUs with CUDA, but in 2024 the project was discontinued. In accordance with the terms of the contract and after receiving permission to publish from an AMD representative, Andrzej opened the code of developments created while working at AMD and allowing CUDA applications to run on top of the ROCm and runtime HIP (Heterogeneous-computing Interface for Portability).
The year before, Andrzej was forced to remove the code from public access after a letter from lawyers made it clear that permission given during email correspondence had no legal force. After this, Andrzej began work on a new edition of ZLUDA, based on the code base that existed before Andrzej started working at AMD. In its current form, development is focused on running applications that use CUDA to accelerate machine learning-related tasks. The project is currently limited to running on AMD GPUs, but will be adapted for Intel GPUs in the future.
The new release adds initial support for running the PhysX physics simulation engine and the Blender 3D modeling system on top of ZLUDA. PhysX support is still in early stages
alpha versions and is limited to the ability to run 32-bit PhysX builds, which allows systems with AMD GPUs to run some older games using this engine. For example, in the game Mafia II, after enabling ZLUDA, it was possible to increase FPS from 26 to 80 frames per second. Implemented texture support sufficient to run Blender using ZLUDA. Significantly improved support for the Windows platform. Expanded capabilities for running frameworks on top of ZLUDA to run large language models (for example, PyTorch) using CUDA optimizations used for NVIDIA GPUs.