Developer Andrzej Janik creates ZLUDA: allowing non-NVIDIA hardware to play CUDA applications

News on February 16th. Although NVIDIA’s CUDA has built solid ecological barriers after years of development, making it difficult for other competitors to break through, developer Andrzej Janik used Intel’s oneAPI to independently develop a new product called "ZLUDA" CUDA compatible solution. The emergence of this solution enabled Intel hardware to run CUDA applications natively. However, the project was later suspended for unknown reasons. The emergence of this project has attracted the attention of the industry because it provides Intel with an opportunity to break through NVIDIA's barriers. However, the reasons for project suspension may be related to technical implementation or commercial considerations. Although it is not yet clear, this suspension may have a certain impact on Intel's development in the GPU field. In any case, the emergence of this project proves that competition in the market still exists, and technological innovation and developer efforts will still promote the development of the industry.

However, ZLUDA’s story does not end here. After receiving support from AMD, the project was restarted and achieved greater breakthroughs. With the introduction of the ZLUDA library, AMD graphics cards can directly run CUDA applications without any code adjustment or migration work. While this does not guarantee that all CUDA applications will run perfectly, it is certainly a significant advance. This innovation is a huge boon for users of AMD graphics cards, who can more easily enjoy the advantages of CUDA applications. In addition, this also gives AMD a favorable position in the competition with NVIDIA, allowing more users to choose AMD products. In general, the birth of ZLUDA brings more choices and convenience to users, and brings more competitive advantages to AMD.

According to the editor’s understanding, unfortunately, AMD also chose to suspend it after supporting this project for two years, casting a shadow on ZLUDA’s prospects. Although it is unclear whether there was pressure from NVIDIA.

Fortunately, the ZLUDA project has not been completely stopped. The project has been made free and open source, so any interested developer can use it and continue research. Additionally, ZLUDA has been confirmed to run CUDA applications seamlessly on AMD’s ROCm platform. For example, developers can use CUDA as the rendering API for Blender 4.0 or V-Ray. Especially for Blender, which natively features the Radeon HIP computational renderer, and now with ZLUDA support, rendering speeds are expected to be even faster. This provides developers with more options to help them perform GPU computing and rendering tasks more efficiently.

In addition, Andrzej Janik is still studying ZLUDA in depth and is committed to enabling AMD graphics cards to run NVIDIA's DLSS technology. Internal tests show that compared to the OpenCL solution, ZLUDA performs better in most cases, sometimes even exceeding 75% performance improvement. However, since the project is still in the development stage, there may be times when it is less efficient. Nonetheless, the emergence of ZLUDA has brought new possibilities and challenges to the CUDA ecosystem.

The above is the detailed content of Developer Andrzej Janik creates ZLUDA: allowing non-NVIDIA hardware to play CUDA applications. For more information, please follow other related articles on the PHP Chinese website!