Home ยป Utilizing the NVIDIA CUDA Library on AMD ROCm: ZLUDA Unleashes Experimental Iterations

Utilizing the NVIDIA CUDA Library on AMD ROCm: ZLUDA Unleashes Experimental Iterations

ZLUDA, an open-source project, aims to implement the CUDA library from NVIDIA using the HIP/ROCm library from AMD as a backend, allowing programs compiled for CUDA to be executed using AMD graphics chips.

The developer of ZLUDA, Andrzej Janik, has been working on this library for several years. Initially, it was developed for CPU integration. However, in 2022, AMD approached Janik to develop the library to support AMD graphics chips. After nearly two years of development, AMD decided to discontinue the project, leading Janik to release it as an open-source project.

Currently, ZLUDA only supports ROCm 5.x, even though AMD has already released ROCm 6. Testing has shown that several programs, which support both AMD and NVIDIA, actually perform better when running through ZLUDA instead of AMD’s own API. Overall, ZLUDA enables the performance of programs that require CUDA on an AMD 7800XT chip, performing on par with the NVIDIA RTX 4070 or RTX 3080.

As of now, the project is licensed under Apache 2.0 as an open-source project. It remains to be seen whether Janik will establish a community to maintain and further develop this library.

TLDR: ZLUDA is an open-source project that enables the execution of CUDA programs on AMD graphics chips using the HIP/ROCm library. Developed by Andrzej Janik, the project was initially meant for CPU integration but later expanded to support AMD graphics chips. Testing has shown that ZLUDA performs better than AMD’s own API for programs supporting both AMD and NVIDIA. The project is currently in the Apache 2.0 license, and its future development depends on the formation of a community.

More Reading

Post navigation

Leave a Comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Advanced LLM AI System by Ollama with AMD Card Support

Unleashing Cutting-Edge Software Execution: ZLUDA Fuses NVIDIA CUDA Library Onto AMD ROCm, Amplifying Test-Driven Versions

NVIDIA Shattering Boundaries: Unveiling the Maverick ARM Architecture PC Processor, a Paradigm Shift in Windows Computing