Getting Almost Constant Karma XPU errors...

   443   8   1
User Avatar
Member
29 posts
Joined: Dec. 2024
Offline
The error I get is that my Nvidia Card fails "Cuda Error". This seems to happen after it renders a few times for single frames.

Have an RTX 4080 Super. Vram is at 7 of 16gigs while Houdini is open. Nvidia drivers are at 591.74.

Any suggestions for diagnosis? Is this a common error?

I don't know how to wheel my nvidia drivers back further than 591 as that's all that is available in my app.
User Avatar
Staff
616 posts
Joined: May 2019
Offline
we've been recommending people to avoid the 590 range (drivers go up in groups of 5, meaning "590 range" is 590-594.99)

If rolling back to a 580 driver is not an option, try a 595 driver and see how you go
User Avatar
Member
8375 posts
Joined: Sept. 2011
Offline
brians
we've been recommending people to avoid the 590 range (drivers go up in groups of 5, meaning "590 range" is 590-594.99)

If rolling back to a 580 driver is not an option, try a 595 driver and see how you go

610 is the current driver train and a new update to it was just released Tuesday. Maybe that one will fair better? The 610 series has been working well so far for me with Blackwell series.

I must say, rolling back to ancient drivers for basic stability is unacceptable and potentially a security risk. There are certain inference models that require the latest drivers so reverting drivers could be a deal breaker for those with mixed use workstations.
User Avatar
Member
29 posts
Joined: Dec. 2024
Offline
Okay, I tried 610 and experienced more stability but had the same problem eventually. I have attempted to render in smaller batches of frames and that seems to help, but it does seem like the more you use Karma, the more it's likely to happen.

Even if I am just using it in the viewport to review what I'm working on with single frames, half resolution, it will eventually trigger.

Requires resetting the computer again.

I am trying 580 now.
User Avatar
Member
29 posts
Joined: Dec. 2024
Offline
For anyone checking this, trying a 580 driver fixed the issue completely.
User Avatar
Member
1065 posts
Joined: April 2008
Offline
stucooke
Okay, I tried 610 and experienced more stability but had the same problem eventually. I have attempted to render in smaller batches of frames and that seems to help, but it does seem like the more you use Karma, the more it's likely to happen.

Even if I am just using it in the viewport to review what I'm working on with single frames, half resolution, it will eventually trigger.

Requires resetting the computer again.

I am trying 580 now.

That is exactly what happens with me on Fedora 44 cinnamon X11 with those drivers.
https://vimeo.com/user2163076 [vimeo.com]
User Avatar
Member
29 posts
Joined: Dec. 2024
Offline
mzigaib
stucooke
Okay, I tried 610 and experienced more stability but had the same problem eventually. I have attempted to render in smaller batches of frames and that seems to help, but it does seem like the more you use Karma, the more it's likely to happen.

Even if I am just using it in the viewport to review what I'm working on with single frames, half resolution, it will eventually trigger.

Requires resetting the computer again.

I am trying 580 now.

That is exactly what happens with me on Fedora 44 cinnamon X11 with those drivers.

You mean that it crashes with those drivers? Or is the video link meant to show?
User Avatar
Member
1065 posts
Joined: April 2008
Offline
With the drivers on Fedora 44 when you use Karma to render for a while just like described above.
https://vimeo.com/user2163076 [vimeo.com]
User Avatar
Staff
616 posts
Joined: May 2019
Offline
stucooke
Okay, I tried 610 and experienced more stability but had the same problem eventually. I have attempted to render in smaller batches of frames and that seems to help, but it does seem like the more you use Karma, the more it's likely to happen.

Even if I am just using it in the viewport to review what I'm working on with single frames, half resolution, it will eventually trigger.

Requires resetting the computer again.

I am trying 580 now.

I know 580 worked for you, but did you have a chance to try 595?
  • Quick Links