Error rendering with Arnold GPU

[am_wilkins]

Hi,

I've been running some render tests with Arnold on the GPU in Solaris, however occasionally my scene starts giving this error and refuses to render when submitted via HUSK.

The machine it renders on drivers are all up to date, latest generation card, frequently does render Arnold GPU frames successfully.

Renders fine locally in the scene.

R 65| 00:00:02  2799MB ERROR   | [gpu] OptiX call failed: (7010) OPTIX_ERROR_DISK_CACHE_INVALID_PATH
R 66| 00:00:02  2799MB ERROR   | [gpu] GPU context creation failed : OPTIX_ERROR_DISK_CACHE_INVALID_PATH

What does this mean, and is there a way to prevent this from happening?

--

Houdini Core Version 20.0.590
Arnold Core: 7.2.5.2

HotA: 6.2.5.2

CPU: AMD Ryzen TR 3990x

RAM: 128GB
GPU: NVIDIA RTX A4000
OS: Windows

[Stephen.Blair · Accepted solution]

Look for something like at the start of the log.

[gpu] setting OptiX disk cache location to: C:\Users\adsk\AppData\Local/NVIDIA/OptixCache

Maybe husk is running as a different user or with different permissions.

The location of the disk cache can be overridden with the environment variable OPTIX_CACHE_PATH. The environment variable takes precedence over this setting.

The default location depends on the operating system:

• Windows: LOCALAPPDATA%\NVIDIA\OptixCache
• Linux: /var/tmp/OptixCache_<username> (or /tmp/OptixCache_<username> if the first choice is not usable), the underscore and username suffix are omitted if the username cannot be obtained

[am_wilkins]

Hi Stephen, thanks for the info.

[gpu] setting OptiX disk cache location to: C:\Users\adsk\AppData\Local/NVIDIA/OptixCache

Unfortunately, running a search for "Optix" and I don't see this line in the logs. 

However, I believe something must've gotten corrupted with the machine so we're either going to try re-installed drivers or re-image the machine.

Running the USD through Husk on a different machine works.

However on this machine also gives I guess the same error with difference wording:

00:00:39  2420MB ERROR   |  [gpu] CUDA call failed : (700) an illegal memory access was encountered
00:00:39  2420MB ERROR   |  [gpu] Exception thrown during GPU execution: an illegal memory access was encountered

PS. It's worth mentioning that the machine is still rendering with Karma XPU just fine.

So this seems an Arnold specific problem.

Will let you know if there is any further errors, info or updates.  

thanks,

amwilkins

[am_wilkins]

Good day,

We've managed to solve this issue and it was in relation to the "APPDATA" path and how our render manager was setting it.
Thanks for the pointers/assistance, etc.

Marking resolved.

amwilkins

[wkyu59875]

May I ask how you solved it? The same problem has arisen now, and only a few projects have such issues. Thank you very much.

[am_wilkins]

In our case it was how we were setting the local "appdata" on our render manager app.
example: LOCALAPPDATA = <USERPROFILE>/AppData/Local

but with normal variable its might be like %USERPROFILE%

[wkyu59875]

My current issue is that the GPU is reporting an error Exception throw during GPU execution: an illegal memory access was counted. Once this error is reported, even if switching to CPU, rendering will not be possible. But reopening the project and trying to delete all materials from the GPU is normal again

[thiago.ize]

 wkyu59875, I think your error is different from what is reported in this ticket (this ticket is about OPTIX_ERROR_DISK_CACHE_INVALID_PATH). What you're reporting is just a generic GPU rendering crash. Make a new post in the Arnold GPU Forum with more details and we might be able to figure out what the bug is and fix it.