CUDA, the famous software environment, is clearly the envy of many. At the same time, with CUDA, NVIDIA has managed to seriously reduce processing times compared with other APIs. It’s only logical that it should attract a lot of interest, and projects such as Zluda have been launched to make hardware other than GeForce compatible!
When AMD financed Zluda!
Initially, Zluda was aimed at Intel hardware and the goal was the same: to make non-NVIDIA hardware CUDA-compatible. But Intel abandoned the project… which was taken over by AMD. Then abandoned just recently. But serious progress has been made, and the project, which is available as OpenSource, is now up and running.
Thanks to Andrez Janik, the man behind it all, we can already run CUDA-enabled applications on AMD graphics cards. Linux specialists Pheronix have carried out some benchmarks. The results show that under Blender, recent AMD cards gain a few seconds. However, the RX 6000 are faster with their dedicated API.
Finally, on Geekbench 5.5.1, the RX 6800 XT sometimes shows very big gains, depending on the scene. For example, on Histogram Equalization, Zluda delivers performance gains of 75%, whereas on Canny, it’s a monumental flop.
In short, with AMD having cut funding, Andrez Janik says he will continue to work on his project. However, he will improve the points that interest him, such as making DLSS compatible with AMD cards.
