I am trying to install the Cuda Toolkit (12.2.0) on an Oracle Linux 8 server with a Quadro GV100 gpu. The driver embedded in the installers are not compatible with the GV100.
If you’re using the runfile installer, you can opt to install just the toolkit, without driver. See here.
Thank you! I will give it a try!
I reinstalled the correct nvidia driver and the cuda toolkit using the run command, deselecting the driver (wrong one). Both finished successfully but the cuda installer dropped a message that the “This installation did not install the CUDA driver …” . When I run nvidia-smi I get the mismatch message with the correct nvidia driver version displayed. “Failed to initialize NVML: Driver/library version mismatch
NVML library version: 550.120” . Any additional advice is greatly appreciated. This is a system that has been running cuda for some time. I feel like I have uninstalled all old remnants but not sure how the previous SA installed (ie, RUN or packagemgr). Thankl you!!!
Sorry I can’t offer much. I’d have expected starting with all Nvidia components removed, that installing the toolkit first, (with or without the included driver) and then installing the driver you want, to end up with the desired result.
Thank you! I just tried again but in the reverse order. If it doesn’t work I will go through the motions the other way around.
Best regards! Gabrielle