Dual GPU VRAM Limits
ebergerly
Posts: 3,255
There is an often-repeated description of how Iray/CUDA/W10 handles dual (non-SLI) GPU's, and it goes something like "The VRAM of the two GPU's isn't stacked (added together), but rather the entire scene must fit into the smallest VRAM GPU".
While that seems mostly true, it implies (and I believe it's been stated here before....) that if your entire scene doesn't fit into the smallest VRAM GPU, then the render will crash to CPU. And based on my tests, that doesn't seem to be the case.
I have a scene that fills just under 20 GB of system RAM, and that gets converted into around 6 GB of GPU VRAM. Since both of my GPU's can handle 6GB (an 8GB GTX-1070 with an 11 GB GTX-1080ti), both GPU's load up to around 6GB and render together.
However, if I use a bigger scene that uses around 30+ GB of system RAM, it starts filling both GPU's, and the 8GB 1070 VRAM stops filling at around 7GB, but the 1080ti keeps filling until around 10GB. And instead of crashing to CPU, the scene renders happily on JUST the 1080ti.
Maybe it was my misunderstanding, but I thought everyone was claiming if it doesn't fit on the smallest you're out of luck and it will render on CPU.
Anyway, it seems that having a smaller VRAM GPU isn't a deal-breaker for larger scenes, it just won't participate in the render. Am I just behind the times on this one, and everyone already knew about it? 
Daz 3D is part of 
Connect
DAZ Productions, Inc.
224 S 200 W, Suite #250
Salt Lake City, UT 84101
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
The GPUs are independent - the idea that the smallest limits all is often repeated but not correct.
That's exactly what I get also - the 980ti vram occasionally fills up and is dropped from rendering, but the 1080ti continues. So the overall render time is increased, but it's still a lot faster than a cpu render.
One thing that I tend to do that often works is stop the render once the 980ti is dropped, ensure the scene is saved and then close and re-start Daz Studio, load the scene and immediately start rendering - a lot of the time the scene then fits into the vram of the smaller card.
Subtle distinction, but important I agree. The basic logic is sound if one assumes that utilizing all GPUs at all times is the goal of the vast majority of people paying top dollar for processors. It saves you time knowing ahead of purchase that a 3gb card wont be usedful for 12gb scenes, even if you buy 4 of them for cheap and stack them together. Not everyone coming into this field understands this at first.
@ebergerly "Anyway, it seems that having a smaller VRAM GPU isn't a deal-breaker for larger scenes, it just won't participate in the render. Am I just behind the times on this one, and everyone already knew about it? "
Maybe not everyone, but it has been discussed and stated many times (almost every thread when someone asking about using multiple GPUs).
...where a lower VRAM card could be useful is if you want better display response than your onboard chipset would offer when dedicating your big VRAM card to rendering only.