AMD Quietly Funded A Drop-In CUDA Implementation Built On ROCm: It's Now Open-Source
Kicking things off was the stable NAMD 2.14 release... NAMD has long offered NVIDIA CUDA optimized builds for this molecular dynamics software albeit only for 2.15 alpha builds is there ROCm support but not for the newer NAMD 3.0 beta builds. But with ZLUDA, you can enjoy NAMD 2.14 CUDA builds accelerated on Radeon GPUs with pretty good performance without any source changes and in fact just using the standard NAMD CUDA release binary.
With Blender the ZLUDA performance was most interesting... Using the CUDA (non-OptiX) back-end showed the Radeon performance to be quite competitive to the NVIDIA cards.
But, yes, after the CUDA back-end was around for years and after dropping OpenCL, Blender did add a Radeon HIP back-end... But the real kicker here is that using ZLUDA + CUDA back-end was slightly faster than the native Radeon HIP backend.
With the more demanding Classroom scene, using ZLUDA with the CUDA back-end on the Radeon RX 6000/7000 series again yielded slightly faster performance than the native Radeon HIP codepath.
It was fascinating to see Blender 4.0 with this real-world, complex software leveraging ZLUDA out-of-the-box for running the CUDA back-end on Radeon GPUs with slightly better performance than its own HIP route.