NVIDIA
Issues with vGPU and cuda
Hello, I have an eval license for GRID and am using XenServer Enterprise edition 7.6 as HV. I have two VMs built and running ubuntu 16.04 Installed nvidia-cuda-toolkit and openssh-server I am running GRID license server off another VM I built two P4-2Q vGPUs on my P4 card Installed the nvidia grid drivers in VMs manually pointing to my kernel source headers Setup nvidia-gridd creating the config file and pointing to license server License seems to be accepted and fps is raised back to normal, X server config says it is licensed. Wrote a couple quick cuda C++ files using gpu, they compile but execute without entering the kernel section (files checked on my dev machine, not an issue with the files as they execute correctly) Reboot machine and I just get black screen (in openxenmanager (linux open source 3rd party xen manager)), using ssh to connect works, but there is no trace of nvidia drivers. I think it may be an issue / conflict between nvidia-cuda-toolkit and the drivers, as I tried installing this last, and before this it seemed to be working fine but after installing it threw "Failed to initialize NVML: Driver/library version mismatch" Mostly just using ssh from console anyway as I can do what I need here, so have not setup vnc viewer or anything similar to connect visually. I assume I will need some driver logs to help narrow this down, am looking for them now Fixed, installed runtime version of cuda tookit
Hello, I have an eval license for GRID and am using XenServer Enterprise edition 7.6 as HV.
I have two VMs built and running ubuntu 16.04
Installed nvidia-cuda-toolkit and openssh-server
I am running GRID license server off another VM
I built two P4-2Q vGPUs on my P4 card
Installed the nvidia grid drivers in VMs manually pointing to my kernel source headers
Setup nvidia-gridd creating the config file and pointing to license server
License seems to be accepted and fps is raised back to normal, X server config says it is licensed.

Wrote a couple quick cuda C++ files using gpu, they compile but execute without entering the kernel section (files checked on my dev machine, not an issue with the files as they execute correctly)

Reboot machine and I just get black screen (in openxenmanager (linux open source 3rd party xen manager)), using ssh to connect works, but there is no trace of nvidia drivers.

I think it may be an issue / conflict between nvidia-cuda-toolkit and the drivers, as I tried installing this last, and before this it seemed to be working fine but after installing it threw
"Failed to initialize NVML: Driver/library version mismatch"

Mostly just using ssh from console anyway as I can do what I need here, so have not setup vnc viewer or anything similar to connect visually.

I assume I will need some driver logs to help narrow this down, am looking for them now

Fixed, installed runtime version of cuda tookit

#1
Posted 02/20/2019 01:03 AM   
Scroll To Top

Add Reply