AMD HIP vs. NVIDIA CUDA vs. NVIDIA OptiX On Blender 3.2
Last week with the release of Blender 3.2 bringing AMD HIP support for Linux to provide for Radeon GPU acceleration, I posted some initial benchmarks of AMD Radeon RX 6000 series with HIP against NVIDIA RTX with OptiX. There was interest by some Phoronix readers in also seeing NVIDIA CUDA results even though OptiX is in good shape with RTX GPUs, so with that here are results of NVIDIA CUDA vs. NVIDIA OptiX vs. AMD HIP with Blender 3.2 on Ubuntu Linux.
Back for my original testing I did carry out the NVIDIA CUDA benchmarks on Blender 3.2 but really weren't too interesting: OptiX support is in good shape with Blender, it was looking at basically "best vs. best" in terms of the optimal GPU support for both AMD and NVIDIA, and the CUDA benchmarks didn't really change the positioning either with still coming out ahead of AMD HIP on Linux. I had meant to just include the OpenBenchmarking.org link to the CUDA results in the original article, but as a few Phoronix readers inquired about CUDA metrics, this article has OptiX vs. CUDA vs. HIP for easy side-by-side analysis.
Even with NVIDIA CUDA up against AMD HIP on Linux, NVIDIA still holds the upper-hand as these results show. As noted in my earlier article, I was unable to get AMD HIP working for RDNA1 or prior GPUs with Blender 3.2 on Linux. These meeting minutes note that they are aware of the RDNA1 "critical bug" and for Blender 3.3 is where they hope to have GFX9/Vega working on both Linux and Windows. Meanwhile for Blender 3.4 is AMD's current planning for having hardware ray-tracing natively supported for Radeon GPUs.
Blender 3.4 is expected to be released in December, so we'll see by them if the AMD HIP ray-tracing support is ready for then or diverted to a Blender release in 2023.
Anyhow, for those wondering how NVIDIA CUDA vs. NVIDIA OptiX vs. AMD HIP stacks up on Linux with the latest drivers on Blender 3.2, here are those benchmarks with the Radeon RX 6000 series and NVIDIA GeForce RTX 30 series graphics cards I have available for testing. GPU power consumption using the exposed power monitoring interfaces under Linux and the exposed GPU core temperatures were also recorded during this Blender 3.2 benchmarking.