I am looking into using SLI (Scalable Link Interface). I have an extra 3090 laying around. I was wondering if anyone knows if VRChat currently has the capability of using such a thing. Particularly with VRWorks - VR SLI by Nvidia.
VRWorks states that “VR SLI for OpenGL is accessed through a new OpenGL extension called “GL_NVX_linked_gpu_multicast” that can be used to greatly improve the speed of HMD rendering. With this extension, it is possible to control multiple GPUs that are in an NVIDIA SLI group with a single OpenGL context to reduce overhead and improve frame rates.”
I am wondering if the application currently supports it. Supposedly it is baked into Unity, so in theory it should support it. However, obviously VRChat has a funny way of doing things sometimes. Would anyone have more insight and knowledge on this?
Edit I also forgot the 3090 primarily utilizes NVLink. Please add that to any insight that anyone may have as well please!
I haven’t seen any mention saying it does support it. I saw a feature request once, not the one linked below, and when I saw it I tried looking up anything recent with that VR SLI, and there isn’t anything recent.
Feature request from 2019, I don’t see any engagement from VRChat staff
If you have the card, nothing to lose by trying it, and maybe a game actually supports it, but it really seems like a dead end
I wish I could say I have nothing to lose by trying it. Sadly, the NVLink bridge is around $260. I would like to know if this would even be relatively beneficial before dropping an extra amount of cash that would likely end up non-refundable.
Yeah, let’s stick with the rest of my comment then, that VR SLI stuff disappeared into the wind before anyone really had a headset, and haven’t been heard from since.
I don’t have an answer to the questions in this thread. However I’d like to note that a single 3090 would probably be more than sufficient alone to run VRChat, as most content is CPU-bound not GPU-bound, and I assume a secondary GPU would be unlikely to improve that by a lot or at all. (Thanks to single-threaded BIRP, complex animators, and unoptimized user generated content with excessive skinned mesh renderers and materials incurring draw calls overhead on the CPU.)