I have a question that came out of a performance test with V-ray for 3d max per CPU.
In these test it shows that performance varies depending on the video card that is installed on your computer.
This test is to show that even a CPU processing, likewise works the GPU (in some proportion to).
It’s like that at some stage in the collection or processing infromacion the same GPU intervenes. Although the engine is working on the CPU.
What I want is to know technically if this is real or due to other factors, such as the PCI bus or the relationship between the bus and reading ram.
I hope I can clarify my doubts.
After participating in a test of rendering in Vray, I wanted to publicly share the results, as there is still the misconception that Vray rendering engine (standard) using only the CPU.
The rendering process involves calculating both the processor (CPU) and the graphics card (GPU), and the amount and type of cores are responsible for executing the calculation process data sent by the GPU, which in turn depends on the number of cores and dememoria type you use. The higher GPU has the largest capacity is the volume of data they send and manage the CPU, which in turn requires RAM depending on the speed and quantity run the calculation.
With this test it was found “and is confirmed” that the type of GPU influences in Vray render time, and that the performance of the GTX 690 (3072 Kepler CUDA Cores - 300W) was higher ligeramnete (by 3.2s) to the GTX 980TI (2816 CUDA Cores MAXWELL - 250W) despite their differences in electrical efficiency; for a third place for the GTX 980 (2048 CUDA Cores MAXWELL - 165W).
1st Place - GTX 690 (4GB - 3072 Kepler CUDA Cores - 300W):
That’s really weird and I wouldn’t expect that.
We have nodes without GPU at all and they render fine. I think the answer has to come from something other than the GPU.
Don’t know what though
The regular V-Ray renderer does not use GPU in anyway at all whatsoever. So I have no idea what those guys tested or how they came to that conclusion. The only thing that might come to mind is updating the VFB while rendering, but it would be extremely weird if any GPU is burdened by that.
in one there’s a 0.7% difference, in the other is 2.1% difference.
If you render the same scene each day you’ll probably see something similar without changing the video card. Our identical farm machines vary in render time by up to 5% rendering the same scene and the same frame - and the machines which do it quicker and slower changes every time.
As Neilg states the differences are very small and could be the result of normal variations between identical renders. Try rendering the same image on the same system 10 times and compare the render times, they will usually be slightly different. If this test would have a bigger sample size, for example 100 runs on gpu1, 100 runs on gpu2 and so on and you would compare the average times for each system they would likely be practically identical. Any remaining difference (I’d seriously doubt there would be any) could be related to inefficient drivers/programs running in the background.