I recently got a RTX 4090 and opened up a scene that I had been working on and noticed that KarmaXPU would constantly fail and display KarmaXPU: Unable to create CUDA context for device 0.

I decided to do some troubleshooting so I made a new scene with just a simple sphere and a grid made into a sweep. Before I created any materials I instantly threw down a karma node in a LOP network and changed it to XPU and it began rendering fine. I made two materials and put them on the sphere and the sweep and then suddenly it errored out the same way it did in my other scene.

I found that the material with transmission set to 1 instantly causes the error. However in my original scene I did not have any transmission materials, only MaterialX materials that used texture files for albedo, roughness, and metal.

[21:07:14] KarmaXPU: device Type:Optix ID:0 has registered a critical error [cudaErrorIllegalAddress], so will now stop functioning. Future error messages will be suppressed [21:08:13] KarmaXPU: Unable to create CUDA context for device 0

Nvidia RTX 4090 Driver Version: 531.18

I've attached the simple scene with sphere and grid that fails and crashes the XPU render and falls back to CPU.

Testing your scene, I receive no errors after browsing to my copy of the artist_workshop.hdr.
I tried turning transmission on and off, and XPU works fine on my RTX3050 8GB running Houdini 19.5.493. nVidia Drivers 516.94

Enivob
Testing your scene, I receive no errors after browsing to my copy of the artist_workshop.hdr.
I tried turning transmission on and off, and XPU works fine on my RTX3050 8GB running Houdini 19.5.493. nVidia Drivers 516.94

Glad to hear!

I rolled back to the Nvidia Studio Driver 528.49 and everything is working. So it seems to be some sort of driver conflict issue.

I had the same issue with my 3060. At some point after adding a few textures it would crash with the same error. rolling back to 528.49 fixed my issues as well.

I'm able to replicate this as well on my end.
So thanks for posting about it.
We'll contact NVidia about it and let you know any updates.
Cheers

Same problem Here with Nvidia RTX 3070 Driver Version: 531.18
rolling back to previous version too

This has been addressed in 19.5.568 (and also backported to 19.0.384)
(ie, after advice from NVidia, we've put a temporary workaround in the XPU code until the driver issue is fixed)
Thanks for your patience guys!

brians
This has been addressed in 19.5.568 (and also backported to 19.0.384)
(ie, after advice from NVidia, we've put a temporary workaround in the XPU code until the driver issue is fixed)
Thanks for your patience guys!

Awesome stuff brians. Always amazed at how awesome everyone at Houdini and everyone in the community is at helping out and solving stuff!

I'm experiencing this same error with Houdini 19.5.569 and Studio Driver 531.61 for the RTX 2080 Super. Any ideas on a stable driver for this? 531.61 is the latest for the 2080 Super here.

brians
I'm able to replicate this as well on my end.
So thanks for posting about it.
We'll contact NVidia about it and let you know any updates.
Cheers

Hi , its happen to me .
In houdini indi when I change from cpu to xpu , with a rtx 4090 with an asset with trasmission and SS .
i have the last nvidia drivers 535.98 i tried with studio drivers too. And the same error appears.

KarmaXPU: device Type:Optix ID:0 has registered a critical error , so will now stop functioning. Future error messages will be suppressed
KarmaXPU: Unable to create CUDA context for device 0
KarmaXPU: Unable to create CUDA context for device 0

tomorrow i will share the hip file if you need it.
thanks!!

ansuter
I'm experiencing this same error with Houdini 19.5.569 and Studio Driver 531.61 for the RTX 2080 Super. Any ideas on a stable driver for this? 531.61 is the latest for the 2080 Super here.

Are you still experiencing issues ansuter?
Its possible I listed the incorrect houdini version. I'm curious if you're still getting issues on 531.61 for houdini versions past 19.5.569

Jackatack
In houdini indi when I change from cpu to xpu , with a rtx 4090 with an asset with trasmission and SS .
i have the last nvidia drivers 535.98 i tried with studio drivers too. And the same error appears.

We've been finding issues with 535.98 (will be getting in touch with NVidia)
What happens if you roll back to 532.03?


thanks!

Brian

ansuter
I'm experiencing this same error with Houdini 19.5.569 and Studio Driver 531.61 for the RTX 2080 Super. Any ideas on a stable driver for this? 531.61 is the latest for the 2080 Super here.

Are you still experiencing issues ansuter?
Its possible I listed the incorrect houdini version. I'm curious if you're still getting issues on 531.61 for houdini versions past 19.5.569

Jackatack
In houdini indi when I change from cpu to xpu , with a rtx 4090 with an asset with trasmission and SS .
i have the last nvidia drivers 535.98 i tried with studio drivers too. And the same error appears.

We've been finding issues with 535.98 (will be getting in touch with NVidia)
What happens if you roll back to 532.03?


thanks!

Hi Brian , nice to meet you.

I reported the bug to support , this is what I discovered debugging the issue.


I can make XPU work again returning my GPU drivers to 531.61-desktop-win10-win11-64bit-international-nsd-dch-whql
So I think it's a nvidia latest drivers problem .


Actually Im having issues with xpu , the same asset , with trasmission and SSS. If I scatter the crystal x 10.000 using an instancer in solaris .Houdini suddenly closes .
Sometimes , an error appear with a problem of memory allocation.
I have windows 10 , 128 gb ram , ryzen 5950x and a rtx 4090.


Now Im installing the new nvidia drivers , i will post when i have more info updates.

Cheers!!


NEWS:
with Houdini 19.5.640 and the lastest studio drivers 535.98 the xpu first problem its solve .No more cuda init errors.


But when I make a renderView with karma xpu with the crystal scattered , houdini closed without an error everytime i try.
If I reduce the intances , it works .

Hi

Jackatack
Actually Im having issues with xpu , the same asset , with trasmission and SSS. If I scatter the crystal x 10.000 using an instancer in solaris .Houdini suddenly closes .

This seems like a different issue
Could you submit a new bug-report with repro scene etc..?

thanks
Brian

brians
Hi

Jackatack
Actually Im having issues with xpu , the same asset , with trasmission and SSS. If I scatter the crystal x 10.000 using an instancer in solaris .Houdini suddenly closes .

This seems like a different issue
Could you submit a new bug-report with repro scene etc..?

thanks
Brian

This weekend I will do it .
Cheers

Hello,

I may have a similar problem.
When I render with KarmaXPU, I got the following message in the terminal.

KarmaXPU: device Type:Optix ID:0 has registered a critical error , so will now stop functioning. Future error messages will Future error messages will be suppressed KarmaXPU: Unable to create CUDA context for device 0

Also, when rendering with KarmaCPU, I got the following message

denoise plugin:.

I am using Houdini FX 19.5.640 on Linux Mint 20.3.
Graphics card---NVIDIA GeForce RTX 4090 with NVIDIA 535.54.03.

PS:I rolled back the NVIDIA driver to NVIDIA 525.125.06 and the problem did not occur.

It may not necessarily be nvidia drivers fault! I've had terrible problems with the latest Linux kernel updates causing problems with (530) nvidia drivers. I still have opencl problems on the CPU but at least it runs on the GPU now.

Shigeru Iriki2
I may have a similar problem.
When I render with KarmaXPU, I got the following message in the terminal.

KarmaXPU: device Type:Optix ID:0 has registered a critical error , so will now stop functioning. Future error messages will Future error messages will be suppressed KarmaXPU: Unable to create CUDA context for device 0


I am using Houdini FX 19.5.640 on Linux Mint 20.3.
Graphics card---NVIDIA GeForce RTX 4090 with NVIDIA 535.54.03.

PS:I rolled back the NVIDIA driver to NVIDIA 525.125.06 and the problem did not occur.

We currently are having trouble with drivers 535.98 and later.
We are in communication with NVidia about it and will let you know when we have any information to share.
In the meantime please roll back to driver 532.03 (or earlier, as you have done)
Thanks for your patience

Hi, do you have cpu with built in GPU? because i have same issue, check GPU device in task manager to know if CPU graphic is on or not. i disabled my own manually

seyed Ali Hossein Izadi
Hi, do you have cpu with built in GPU? because i have same issue, check GPU device in task manager to know if CPU graphic is on or not. i disabled my own manually

No , my cpu is a ryzen 5950x without gpu.

Hi guys

More info on this:
NVidia have found the bug and will be rolling out a fix in the coming weeks.
Until then, you can use this environment variable if you're having driver issues.


It should be in H19.5 from about 19.5.697