Barefeats has once again put a short DaVinci Resolve benchmark online, which can provide some interesting information about the GPU substructure of the popular editing, grading, audio and compositing program. Barefeats only tests on the Mac, but there also with the old Mac Pros (the "cheese graters"), which can still record Nvidia cards via PCIe.
The results were recorded under Resolve 14, with version 15 benchmarks to be submitted as soon as a final version is available. The results tend to coincide with our experience, even if barefeats only uses a single Resolve function (Noise Reduction) as a benchmark.
Only with an Nvidia card you can use all three GPU libraries under MacOS, whereby CUDA still calculates fastest under Resolve. OpenCL renders about 10 percent slower under Nvidia, Metal is about 15 percent behind CUDA.
When using AMD graphics cards (in this test only VEGA GPUs) OpenCL with up to 50 percent performance advantage can sometimes clearly distinguish itself from Metal.
Actually Metal should deliver the best results, because this is the most modern GPU interface, which is just advertising with its efficiency through low overhead. The results are therefore surprising and can now mean two things. Either the CUDA and OpenCL effects are even better optimized and the metal implementation must still mature in the future. Or metal is not so suitable for use as a video effect accelerator.
However, the (developer page of Apple), which explicitly shows DaVinci Resolve as a demonstration image for advertising purposes (and curiously not its own FinalCutProX), speaks against the latter.
And another interesting side information to take home from the Bearefeats test: An AMD Radeon Vega64 plays in the same league as the Nvidia Titan XP under Resolve.
Translated with www.DeepL.com/Translator
