Run NVIDIA CUDA Apps on AMD GPUs with This Project

NVIDIA's CUDA development platform is the most popular among its competitors. While AMD has ROCm, it can't reach CUDA's level of fame. But did you know that NVIDIA's main GPU rival was once trying to bring CUDA to its Radeon cards? Now, you can enjoy what's come out of its funding.

Meet ZLUDA, a project by Andrzej Janik, who was first developing it while working at Intel and then employed at AMD. He was supposed to adapt ZLUDA for use on AMD GPUs with HIP/ROCm, according to Phoronix, however, the company discontinued its funding this year.

Eventually, Janik released ZLUDA as an open-sourced project, and although you should understand that it's in alpha now, it works pretty well, as proven by several outlets. Phoronix has tested ZLUDA and concluded that CUDA-enabled software does run atop ROCm without any changes.

As you can see from its experiments, Phoronix tried the Classroom benchmark for Blender and found that the latter produced slightly better results: Radeon RX 7900 XTX rendered the scene in 20.89 seconds on HIP and 18.44 seconds on ZLUDA. Similar results were achieved with other GPUs.

Image credit: Phoronix

However, it's not always better to use ZLUDA, some Geekbench tests, as pointed out by PC Gamer, showed it's sometimes slower than OpenCL.

While Janik doesn't have AMD's funding now, the development might slow down. Hopefully, making the tool open-sourced will bring more helping hands that can move the project forward.

Check ZLUDA out here and join our 80 Level Talent platform and our Telegram channel, follow us on Instagram, Twitter, and LinkedIn, where we share breakdowns, the latest news, awesome artworks, and more.