This is quite amazing to me that my GPU is actually slower than my CPU. i have a rather complex scene I am rendering that is about 25 minutes per frame. I know it is a lot but it is a complex scene with linked and appended items. The GPU render is 25 minutes and the CPU shows 24 minutes for a render.
I tried using a simple cube with fur to give it something to work on. The GPU did respond faster . I am not sure if the reason why my complex scene with linked and appended items are making the GPU the same speed as the CPU.
In a similar scene with an additionally added linked object, the PU will not even render. It shows it is out of memory but if I switch to the CPU, it will render albeit slow. Any ideas what I am missing if anything?
I have tried all tile adjustments and worked with sampling and compression as well and still the same timing results. With the complex scene the GPU is the same speed as the CPU.
But to clear up doubts, you perform the following benchmarks with official Blender 2.79 (2.79b) and compare results with others. You must download the corresponding files of each thread:
Sometimes with greater complexity, the GPU will run slower than CPU. Many of the GPUs resources are taken up when using CUDA as opposed to OpenCL, which is why you are also seeing the out of memory error. CPUs are more scalable and have access to more system resources, like system memory and swapping, so doesn’t suffer the same problems unless you have one hell of a scene.
My CPU renders faster than my GPU, but that’s because I’m using a Ryzen 1700 with a GTX 970. In some scenes this can be seconds to minutes faster. On more simple scenes, the GPU wins, but the more stuff I throw at it, the CPU starts to pull ahead.
Multi GPU systems should perform better, as long as resources are available.
You wrote in the GPU benchmark thread that that was the time obtained with GPU, right?
What time did you get with the CPU benchmark file? (using CPU of course)
Please, not with the same GPU file. The CPU benchmark must be done with the file downloaded from the CPU benchmark thread. It is the same scene, but with tile size adjusted for CPU.
Thanks for reminding me to use the CPU version. With the CPU benchmark file the time is:
5:19.20 (16 tiles)
5:19.00 (64 tiles)
Windows 10
32 gig RAM
17-7700 3.6 GHz
Not sure how that compares to others but those are my stats. What ajm said about how the GPU uses resources make a lot of sense in how my computer is working. I guess if I would have had dual CPU’s maybe it would have changed things. The GPU was faster running the GPU benchmark test.