Hello everyone, I tested my code on Tesla A100 and Tesla A40, my code will create ~1000 traversable pipelines, and all things are good on Tesla A100, while things go wrong on Tesla A40 as the figure shown. I check the VRAM, GPU exclusive mode, … and can not figure out the reason, I wonder what will cause this fail and how can I solve this on Tesla A40.
Did you follow the advices inside the error message and deleted or renamed the file?
You probably need to add some more information.
Are these two different systems?
Are these different users?
Is the user name part of that optix7cache.db path?
On the failing machine, does the user match that name and have write access rights to that optix7cache.db file?
Did you set the cache location explicitly inside your application?
Please read this chapter about compilation caching inside the OptiX 7 Programming Guide:
https://raytracing-docs.nvidia.com/optix7/guide/index.html#context#compilation-caching
1 Like
Okay, there are indeed some permission issues on these two nodes. These two nodes share the same file system thus exactly the same system, while I can successfully run ncu on A100 node, while failed on A40 node due to some /tmp/lock file. I will check the doc, thanks.