Running out of Vram with RTX 5000 on Centos 7.8

[apps_wefx]

Hello!  We are running SP 6.2.2 on Centos 7.8 using RTX 5000 cards (16gb vram).  Driver is 450.51.06. 

We are finding that any asset we bring into Substance Painter will eventually trigger a 90% GPU ram warning at the OS level, and things will start to get slow. 

We are using Teradici software pcopip client - which may be taking some GPU resources, but this should be relatively minimal. 

Any suggestions on how to avoid this issue? Thanks greatly.

[Jeremie Noguer]

That seems about right, the log indicates around 14GB of textures stored in vram. This can go up pretty quick with a UDIM project. Each layer for each channel is raw texture data, sometimes in floating point, that's a whole lot of data that we try to keep in vram cache as much as possible. As a general rule, Painter will use as much vram as is available until it runs out and start caching in ram and disk.
I would assume this is a somewhat advanced layer stack though. If you get this kind of usage with a very simple or empty stack, then there is an issue.

[apps_wefx]

Thanks Jeremie,

Yes, this is lots of udims and lots of layers.  We did notice that things seem faster in windows on the same machine.  I also noticed that you recommended disabling Vsync and Threaded optimization - but those settings are not exposed in the Linux Nvidia Xserver control panel.  So if there are any tips for optimizing SP in Linux it would be appreciated.

So assuming our artist are pushing the limit (which I certainly can't complain about) - are there techniques for keeping things fluid and avoiding hitting the swap point?

Many thanks!