VMware vRealize Operations Nvidia Management Pack
For users who have been using Nvidia GPU for machine learning processing and huge data processing, many a times, you like to know how is the GPU card been utilize and if its is sufficient.
If you are running VMware platform you will be in luck. vRealize Operations (vROps) has just the management pack from Nvidia can help you on that.
If you are using GPU on your VM and you are not using Nvidia GRID (aka Nvidia AI Enterprise - NAIE) technology but using a passthrough, you might want to explore of Nvidia GRID can meet your requirement. Only certain application required the entire GPU card that is when you use passthrough. However, if that is not the case, you might have over provision your card and might be wasting resource that can be use by other.
To give you a quick explanation, Nvidia GRID was the technology that is introduced by Nvidia and supported by VMware vSphere to slice your GPU just like how you do it on CPU with partnership between the two companies. With GPU sharing, you are still using the native Nvidia drivers and capability and not losing any of it. Unlike some other methods would be doing emulation. This give users the benefit of cost saving since GPU card are not really that cheap. With the support, vSphere is able to also perform vMotion between ESXi hosts that has GPU card. This is truely magnificent. Check out the VMware doc and Nvidia video.
Let's head back to vROps. In order to monitor the usage of the GPU using Nvidia NAIE, you will need to first have vROps in your environment. You should have this if you are not as you will definitely see the benefit of vROps. Next to extend the capability, you can always install management pack. In order to search for the management pack, you will head to the VMware marketplace. Do take note of your vROps edition as some management pack only support a specific edition or later.
Below are some of the resources if you are looking at Nvidia management pack and as well to get the latest version.
VMware Marketplace
https://marketplace.cloud.vmware.com/services/details/nvidia-virtual-gpu-management-pack-for-vrealize-operations-1-0/?slug=true
However, sometimes vendors, like Nvidia do release new one rapidly and have not submitted the new releases. You can still check out the vendor website for the software version available.
Nvidia GRID Software Version
https://docs.nvidia.com/grid/
At point of writing, the latest release is 2.0 and the release notes is as below with the supported vROps version and Nvidia drivers.
Nvidia Management Packs Release
https://docs.nvidia.com/grid/vrops/2.0/grid-management-pack-vmware-vrops-release-notes/index.html
Comments