Blender 3.4 HIP Performance With Radeon RX 7900 Series + RDNA3 OpenCL Compute Benchmarks

Written by Michael Larabel in Graphics Cards on 15 December 2022 at 09:30 AM EST. Page 2 of 6. 40 Comments.
Radeon RX 7900 XT/XTX vs. NVIDIA OpenCL Linux Compute

Starting off with the very basic and well known BMW scene, the Radeon RX 7900 series performance with Blender HIP shows much better performance than the RDNA2 hardware I had available for testing, but still is slower than NVIDIA's RTX 3080/3090 series even with just the CUDA back-end. Using NVIDIA's OptiX back-end easily still beats the AMD HIP performance by wide margins thanks to the RT cores. In 2023 AMD is hoping to have their HIP ray-traced back-end support merged into Blender so hopefully at that time we'll be seeing much better performance on the Radeon side for this open-source 3D modeling software.

Radeon RX 7900 XT/XTX vs. NVIDIA OpenCL Linux Compute

The Classroom scene is more demanding than BMW and thus a more interesting test target. Here the Radeon RX 7900 series GPUs were at least competing with the NVIDIA GeForce RTX 3080 and its CUDA back-end, but obviously still not running up against the OptiX performance. At least using the Radeon RX 7900 series with HIP for rendering these Blender scenes is still much faster than CPU-based rendering.

Radeon RX 7900 XT/XTX vs. NVIDIA OpenCL Linux Compute

While rendering the classroom scene, the Radeon RX 7900 series were consuming less power than the GeForce RTX 3080 Ti and RTX 3090.

Radeon RX 7900 XT/XTX vs. NVIDIA OpenCL Linux Compute

Related Articles