After completing Stage XXIV: Infrastructure Upgrades it was time to invest again…..
I used the Nvidia K2 GPU since several months inside my datacenter. When Nvidia made Grid 5.0 available an existing tiny little GPU, which would be perfect for homelabs, was announced to support the vGPU feature: The P4
I was able to get two of these single slot cards for my 24/7 gear running at my main datacenter at my house.
This card is designed for Deep Learning, but now works perfectly for GPU Virtualization. It only occupies one PCIe x16 slot and does not require a dedicated power connector. Maximum power consumption is 75W.
These are the supported vGPU profiles:
After installing the Nvidia P4 physically inside my two IBM x3650M4 vSAN All Flash servers, I installed the Grid 5.0 Host Extension via Update Manager and rebooted the machines.
I modified the Host Graphics configuration and the P4 settings within ESXi:
One special configuration is needed to use the Pascal based GPUs for vGPU:
ECC must be disabled!
After these configuration steps, I configured the first VM to use the new P4 card:
The VM booted up, I installed the Nvidia drivers and ….. got no GPU acceleration at all…..
You must configure the VM to use the previously installed Nvidia License Server!
First you need to upload your Nvidia licenses to your server:
You must configure each VDI or RDSH Server to use the license server:
And after that configuration, the license is requested from the server and your vGPU is working. Depending on your vGPU profile and your configured resolution one of the following licenses is required:
You can monitor the requested licenses at your license server overview:
You can easily monitor the GPU parameters using the CLI within ESXi:
Another very useful feature within the Grid 5.x release is the vRealize Operations Manager vGPU Management Pack.
This solution allows you to monitor all your Parameters from the vROPs Dashboard:
Stay tuned for the next Stage XXVI: vSAN All Flash with NVMe