Hi there,
My current workstation comprises of 2x Xeon E5-2699 v3 + Quadro K6000
I have always avoided GPU rendering as my CPU’s are really quick and in my current setup much faster than using CUDA, even with Hybrid rendering.
I am contemplating getting 2x RTX 2080 TI’s to test for GPU rendering and am curious as to what performance I should expect, compared to my CPU’s?
Thanks
We have these CPUs at office, 2 of them score around 220 in the new Vray bench
A single RTX 2080Ti is nearly twice as fast..scores around 380 in Vray Bench
considering you are keeping you cards cool and you are not throttling… This is a massive upgrade in rendering speed over your CPUs
Thanks for the reply.
I have 2x 2080 TI’s arriving today so I can test and compare. Looking forward to the results 
show us the results you´ll find!
my big issue with that the rtx still have low ram…most of the scenes we work with exceed 12 Gb of ram constantly I cannot see this working with gpu for us any time soon.
nvlink? titan rtx?
ive found my issues with gpu are not usually ram based, it seems a scene that would take 30+ gb to render on cpu happily renders on my 12 gb titan (while using more than 30 gb of cpu ram too!) it does a good job offloading textures etc.
my issues are the few critical low level settings which i use on every project - affect all channels - trace sets.. multimattes in certain matrerial combinations, a few others.. i often start a job on gpu and 90% of the time i end up switching to cpu halfway through.
I have both of the 2080 TI’s installed now and have been doing some testing on existing scenes.
I am pretty impressed with the speed, CUDA bucket mode is pound for pound quicker than the CPUs in production mode but what I am finding is that quite a few key features of VRay which are an integral part of our existing workflow / pipeline are not supported on the GPU (material select render element, trace sets, etc)
For us, it would take a bit of an overhaul to start going down the GPU route, which isn’t out of the question if the payoff is worth it. It would just be a much easier transition if more of the integral features were supported in GPU.
Yeah those are very very expensive investments, not sure if they are justified yet - they will be in the future though for sure.
The issue is right now gpu uses various techniques to fit the scene into memory. This is also true for cpu. For example I have machines with 48-128 gb of ram and few with 16 gb of ram. A scene that loads into 30 gb of ram on large ram machines also renders on machine with 16 gb, though a bit slower (this is cpu). I think this affects the gpu performance though a bit and kind of diminishes its true potential. Additionally what gpu is good at (smaller scale renders) aren’t that difficult for modern cpus either. So its not a directly linear progression where gpu is that much more faster and completely overtakes the leading cpus.
how are you going to cool the cards..water or air?
it really depends on your scenes. heavy arch vis / vfx fails miserably on GPU.
most of my scenes are 32gb plus in ram usage and just wont render meaning CPUs are still important for the time being. if you are doing that C4D style spheres and a few cubes then GPU rendering will be very fast though and you can join the people on instagram doing those funny late capitalist daily images.
i hope in a couple of years the vram limitation can be over come, cos things are just getting more and more complex. i do note that redshift does a much better job with heavy scenes and anything with volumetrics than vray GPU and is quite cheap