Questions about multiple GPU rig

RobertDy

0 0

I just ordered a new workstation with a RTX 3090 and RTX 2080 Ti (the latter was with me for some time, I intend to put it in the new rig). The rig hasn't arrived yet, and I'm burning with curiosity and excitement...Assuming the rig is optimal for both GPUs performance, I'm curious to know:

1) is there some kind of formula for CUDA core number to render speed? for example, both GPUs mentioned will have 3 times more CUDA cores than my existing RTX 2080 Ti, so should I expect the render speed to increase by thrice (or more or less, if there is such a 'formula'?)

2) I read that the assumption that VRAM-wise, the scene will fall back on the GPU with the lowest VRAM, but this assumption is incorrect. In any case can I expect the 3090's higher VRAM to speed up the scene's navigation and such?

3) do I need to do any software configurations to enable and optimize the performance of both? I read in a post years ago that DAZ doesn't like SLI but I'm not sure if this is still relevant.

Thanks!

Richard Haseltine

1) not really - certainly not between generations, and the RTX adds it own cores for ray-tracing to the mix which will have an effect on many scenes but how great will depend on the nature of the scene.

2) I'm not sure what you mean - given two cards of different capacities iray will use both if it can, but if not it can drop the lower capacity card  and continue to use the higher capacity card. Memory doesn't impact speed, as long as it doesn't run out

3) No, just make sure your driver is new enough for the version of Iray you are suing. SLI is not supported by iray, nut nVlink is - however, you need to two 3090s or better to use that (and it is used to pool material memory, but not other data, and there is a performance hit)

RobertDy

0 0

Thanks Richard, your info is always of great help. About 2), I was referring to how smooth the viewport navigation is or the absence of 'clunkiness', I've always had  the impression that this is influenced by VRAM, do correct me if I'm wrong.

Another thing: do multiple GPUs speed up dforce simulations?

AgitatedRiot

Richard Haseltine said:

1) not really - certainly not between generations, and the RTX adds it own cores for ray-tracing to the mix which will have an effect on many scenes but how great will depend on the nature of the scene.

2) I'm not sure what you mean - given two cards of different capacities iray will use both if it can, but if not it can drop the lower capacity card  and continue to use the higher capacity card. Memory doesn't impact speed, as long as it doesn't run out

3) No, just make sure your driver is new enough for the version of Iray you are suing. SLI is not supported by iray, nut nVlink is - however, you need to two 3090s or better to use that (and it is used to pool material memory, but not other data, and there is a performance hit)

PCI Express lanes are often distributed across two slots, meaning that the second PCIe port may only run eight lanes electrically. This also means that the speed of one physical x16 PCIe port decreases when you use expand out to a second slot.