you recommend to use 2xRTX3090 graphics cards for Omniverse Create XR. Is it because of SLI over the NVLink? Does it make sense to use 1xRTX4090 plus 1xRTX3090(which we already have)? And in which way can it be usefull. Can one card compute PhysX while the other is computing real-time rendering? Or is there any other way to utilize this GPUs combination. What will be the benefit?
We do not recommend using SLI with Create or Create XR. Both software apps will correctly find, use and take advantage of your multiple GPUs without needing to physically connect them in SLI.
You can use NVLINK if you wish but this is for Quadro cards only. Like 2 x Quadro A6000 cards.
As a good general rule you want to always have your fastest card as your primary, so your applications run faster.
Create and Create XR will use all available GPU power very well, so if you upgraded to one 4090, and one 3090, in theory, that would be a big speed advantage over 2 x 3090s. HOWEVER, as yet we have not tested a mix of a 4090 with a 3090 for stability.
I haven’t had a chance to install both of them. They don’t physically fit in the case using the motherboard that I have. I’m searching for water cooling system to be able to install both. Water blocks are thinner that heat-sinks.
Just to clarify on the use of VRAM and Create. Create cannot “combine” two seperate cards VRAM into one pool. Create, like most GPU software can only use the amount of VRAM on the smallest card you wish to use. So if you had a 24GB and an 8GB card, Create can only use a maximum of 8GB. All cards in use are bound by the smallest amount. So with a 24GB 3090 and a 24GB 4090, it would only use a maximum of 24GB.
And as a follow up to that, in general it is not a good idea to mix cards. Ideally you want the same card repeated. Certainly not different memory amounts or GPU classes. We do not gaurantee support and stability with mixing a 3090 with a 4090 as they are very different cards.
Yes correct. NVLINK is not supported by Create or by the Geforce series. So even with two cards you will only get a maximum of 24GB. But ofcourse with double the cards, you get double the speed of rendering.
Just an update that I’m currently testing this setup with TRX40 system, 3090 and 4090. And, for example, on my scene the rendering time is dropped from 40 sec/frame (on 4090 only) to 21sec/frame in Path-Trace mode.
Since (as far as I understood) you still have to add “–/renderer/multiGpu/enabled=true” command in 2022.3.1 argument to the Create’s .bat file shortcut for multi GPU support, the command line reads: “Disabling DLSS Frame Generation for at least one view due to incompatible render outputs” - so, i guess, the DLSS stuff does not quite work with this setup.
Yes that makes a lot of sense on the DLSS support. That is difficult across two different cards. Multi-GPU should be ON by default. You should not have to add a command line. Can you try without the extra command line and let me know.
Actually, I have just found out that in fact you are correct to add the manual code. For two or more cards are the same specs, the OV launcher will auto detect multi GPU. However, when you have two different cards, and in your case, two different classes, it will not auto detect. You can turn it on and off manually inside Create, in the Preferences.