I’ve encountered an issue while utilizing the GPU rendering mode where I’ve noticed a significant delay in rendering the empty black areas of my scenes. This is in contrast to the CPU rendering mode, which resolves these areas almost instantaneously. Considering there is nothing to render in these empty spaces, this behavior strikes me as unusual. Could this be a bug, or is it an expected behavior within the GPU rendering process? Any insights or suggestions to improve rendering efficiency in these areas would be greatly appreciated.
From what I can observe there is nothing to worry about. The CPU buckets are smaller which would result in faster calculations of their area whereas the RTX buckets are double the size. That is why it takes slightly longer to calculate.
What I would encourage you to do is compare Render Time and share it with me. However, even with Render Time, please, keep in mind that CPU and GPU are completely separate Render Engines that work with and support features differently.
I also always found it strange, that Vray GPU would take rather long on empty buckets. And without knowing the specs of the machine above, it also seems to be slower than the CPU renderer (because of that) in this particular case.
I’d like to share the results of my latest tests:
I’ve tried rendering a completely empty scene, except for an invisible area light. I’m using a resolution of 3840x2160px and the default settings. In CPU mode, it takes the same amount of time whether the light is on or off, about 0.5s. In GPU mode, both RTX and CUDA, with the light off it takes 2.3s, and with the light on it takes 10.9s.
I’ve noticed that the moment a light is introduced into the scene in GPU mode, rendering empty areas, for some reason, slows down significantly. Moreover, even with the light turned off or completely removed from the scene, GPU mode still takes more than three times longer than CPU mode (2.3s vs 0.5s). It seems like something is going on here, doesn’t it?
I’m investigating this issue now, there seems to be a delay on the empty pixels in CUDA/RTX engines compared to the standard V-Ray CPU renderer
I will see what we can do about that
For comparing the performance between V-Ray and V-Ray GPU, best to use scenes that take minutes rather than seconds to render.
Edit, please use Progressive Sampler with the GPU engine for the time being, it works well on GPU. The devs will take a look at this when comes a chance
