Blender Cycles Render Engine Benchmarks With NVIDIA CUDA On Linux

Here is a look at the performance of the Blender 3D modeling/creation software with its Cycles Engine when making use of NVIDIA's CUDA API for GPU acceleration. Tests for this initial comparison include NVIDIA GeForce GTX 1000 "Pascal" and GTX 900 "Maxwell" graphics cards.

The latest test profile available via OpenBenchmarking.org is support for Blender! You can now run a fully-automated OpenCL/CUDA Blender Cycles benchmark with various Blender reference files via a simple phoronix-test-suite benchmark blender command for it to be fully-automated in a reproducible manner from start to finish. The test profile currently makes use of Blender 2.77a plus also ships with Pascal support manually added in for the time being.

The Blender Cycles scenes used for rendering are those that were offered up earlier this year via the Blender Developer Blog. Also, thanks to Render.st for bringing it to my attention that Blender can now be properly automated for benchmarking and their help in getting this test profile established.

For your viewing pleasure, the initial Blender CUDA benchmarks to share are from various NVIDIA GPUs on Ubuntu 16.04 Linux with the 367.27 driver and CUDA 8 RC. These cards included the GeForce GTX 1080, GTX 1070, GTX TITAN X, GTX 980 Ti, GTX 980, and GTX 960. In a follow-up article I'll be trying out the Blender OpenCL support with the AMDGPU-PRO stack to see how that goes and will post some benchmarks, including from the new Radeon RX 480 graphics card.

In addition to the raw results for this article, there are also power metrics too for additional reference points. All of this Blender benchmarking was done in a fully-automated and standardized manner via the open-source Phoronix Test Suite benchmarking software. Enjoy all these interesting benchmarks and the work we do on benchmark automation? Consider joining Phoronix Premium.