I added a GEForce GTX 1650 to my PowerEdge KVM Host. I want to consume the resource and attach it to a K8s worker node

Hi there. I have been appreciating and using MAAS for over a year. And wow. Thanks!

I just got a small upgrade for my homelab and now I intend to use a GPU for my Folding@Home with my large server that normally runs the K8s Core juju default deployment with some buffs (larger workers and RBAC).

I received this little consumer-grade gpu in the mail today, popped the chassis on the PowerEdge and plunked in the upgrade.

Then I hastily destroyed my old Core k8s deployment, deleted my KVM host off MAAS, re-initialized the RAID array on the PowerEdge, re-enlisted it in MAAS, re-commissioned it, and re-deployed Bionic to it, again making it a KVM host.

Now I want to…

  • First make sure that the deployed KVM host has the hardware working on board (how do I even do this? ssh in and try lspci? I’m clueless here…)
  • Second, maybe set up a tag for passthrough to vms using something like this? : https://ubuntu.com/blog/hardware-discovery-and-kernel-auto-configuration-in-maas … BTW it looks a bit arcane to me to start. I am going to need to adapt it to not target a Tesla because I have a vastly underpowered sibling of the Tesla. I can probably figure out the part to match the card using xpath, but I have nary a clue as to how to find out what kernel opts to use. (I suppose I can look on StackExchange somewhere to begin to get a clue.)

If anybody thinks they can help be hack through this, I’ll be super grateful.

Let’s fight Covid with FoldingAtHome :slight_smile:



I cross posted here with my latest updates:

1 Like

@db0west, did you ever get closure on this?