2024-02-14 15:08:20
AMD has introduced an interesting solution based on ROCm technology to enable NVIDIA CUDA-based applications to run in Radeon graphics cards without the need to make any additional modifications to the source code.
AMD has decided to sponsor this open source project, known as ZLUDA, for the past two years. The initial goal of the project was to make the CUDA API compatible with Intel cards, but the desire of developer Andrej Janik, whose services AMD hired in 2022, was to redesign the project to support Also AMD Radeon cards in the HIP/ROCm platform.
The project took two years to develop, and now allows NVIDIA CUDA-based applications to be compatible with AMD cards without the need to modify the source code.
What is noteworthy is that this project was secretly funded by AMD, which may open the door for Nvidia to file a lawsuit against AMD because of this support, and there is a lot of talk revolving around this matter, especially the project’s continued compatibility with the ROCm 5.7 definition, which raises questions about AMD’s involvement. Continuous.
Performance tests revealed by the site showed Phoronix AMD’s ZLUDA library performs well in various tests. ZLUDA allows NVIDIA CUDA applications to run directly on AMD Radeon cards without any change in the source code.
Using the ZLUDA library, developers can effectively bypass NVIDIA CUDA limitations, allowing Radeon cards to run without issues. Although AMD’s support for the project has ended, it remains open source with ongoing contributions from the developer community.
Benchmarks reveal that the ZLUDA library effectively improves performance in certain applications such as Blender 4.0, allowing Radeon cards to outperform the HIP/ROCm platform by up to 10 and 20%. With the Geekbench test, the results revealed that the ZLUDA library provides up to 75% better performance compared to the OpenCL interface.
Related articles
1707927124
#AMD #developing #ROCm #technology #support #NVIDIA #CUDA #applications #Radeon #cards