It’s been a few weeks since Blender 2.72a was released, so it was about time to test it out and see how CPU and GPU rendering speed perform in the latest Blender version.
As in the last tests, we are using for benchmarking the Pabellon Barcelona Cycles scene, that Hamza Cheggour published on his eMirage site. As soon as the rendering was finished, I was amazed by the results. In this version, CPU rendering is faster than GPU. I repeated the test, to make sure there was no mistake, and tried to understand why this happens.
The first thing to notice is that the GPU rendering time increased a bit in this version. This is caused by the migration to CUDA 6.5. Blender started using cuda 6.5 several versions ago, but this is the first version that does not work with an older version of cuda.
The other thing that explains the sudden shift in CPU vs. GPU rendering speed is the further optimization done on the CPU rendering code in Cycles. This, coupled with the fact that RenderStreet is using dual Xeon servers (32 threads), makes the CPU rendering faster on this version of Blender.
Also check out our first test conclusions on the previous Cycles releases, where we analyse how older Blender versions perform.
In my opinion, this speed difference favoring the CPU will differ from project to project. I expect some projects to still retain better the GPU speed, while complex project to work faster on CPU as the one in our benchmark.
How is 2.72a version working for you? Are you seeing the same improvements in CPU rendering speed? Let us know in the comments.
Sorin Vinatoru
CTO RenderStreet
