(ANSWERED) Karma XPU NVLink support?

   1888   5   0
User Avatar
Member
250 posts
Joined: June 2016
Offline
Hello everyone!

I am wondering if Karma XPU supports NVLink, or if there is a plan to support NVLink, I am asking because the new RTX 6000 ADA dropped NVLink support and still only boost 48 GB of VRAM, so it seems the only way to get more than 48 GB for GPU rendering is through NVLink, hence the question.

Thanks!
Edited by GCharb - March 17, 2023 07:05:48
User Avatar
Member
44 posts
Joined: March 2023
Offline
I`m pretty sure there is a forum post on this subject somewhere already.
I`m pretty sure its automatical and the renderer doesnt see the diference if evrything is propperly installed,
Although i Could be wrong.

All I know for sure is that in rhel you need to enable to add /fm on your modularity-stream when installing the drivers.

https://developer.nvidia.com/blog/streamlining-nvidia-driver-deployment-on-rhel-8-with-modularity-streams/ [developer.nvidia.com]
User Avatar
Member
250 posts
Joined: June 2016
Offline
nicolaas.tanghe
I`m pretty sure there is a forum post on this subject somewhere already.

Hello, and thanks for your reply!

I made a search for NVLing before posting, and the search returned nothing, hence my post, and from what I read you need a render engine that supports NVLink for it to work, but maybe it is only the case with gaming cards, and not workstation cards??? 🤔

Here is a link to such an article!

Should you use NVLink or SLI or multiple GPUs for rendering? [irendering.net]
Edited by GCharb - March 16, 2023 18:52:30
User Avatar
Staff
470 posts
Joined: May 2019
Offline
NVLink is not automatic, but its not too hard to get going in code.

The main problem is that it does not support a bunch of the hardware features we use in XPU (ie RTX data, sparse textures, etc...), so its a no-go for XPU for the foreseeable future I'm afraid.
User Avatar
Member
250 posts
Joined: June 2016
Offline
Thanks for the reply Brians, that's a bit of a bummer, XPU is real good, and fast, but VRAM limitation is a problem with it, but if that's what it is, not much we can do!
Edited by GCharb - March 17, 2023 10:10:21
User Avatar
Member
85 posts
Joined:
Offline
brians
NVLink is not automatic, but its not too hard to get going in code.

The main problem is that it does not support a bunch of the hardware features we use in XPU (ie RTX data, sparse textures, etc...), so its a no-go for XPU for the foreseeable future I'm afraid.
I really hope that you at least support "out of core" because if not their is no point for using xpu in VFX projects.
  • Quick Links