A friend in a different support group posted this. This is what I have been observing for a few years. Its nice to see some recent studies.
https://www.journal-of-hepatology.eu/article/S0168-8278(26)00018-8/fulltext
NOTE!: This is NOT an excuse to keep drinking! You degenerate recovering alcoholics, lol(like me...)
My background: I followed DigitalSpaceport's 24gb home AI server(dual rtx3060s) build but with an dual processor hp640(build used hp440). I installed nvidia driver on the host and lxc container, no problems. Super fun, both GPU's passed to LXC! Just messing around with llama3.2-vision I was getting 57/s tokens, and that's with V3 Xeon cpu's. Just upgrading to V4 will improve my memory speed.
So now I want to mess with some video gen with VM's I know I can't combine GPUs for video generation(like I did in llama language models...) so I wanted to create two VMs, and pass each GPU to each... then just do parallel seperate encoding with the two VMs... Well, the host keeps claiming one of the GPUs and it something stoopid simple... I've only done the dual gpu passthrough once with the digitalspaceport guide so I'm new to this. I have added some screenshots of commands, and there's one screen of my previous successful lxc dual passthrough too.
The solution is something simple and its driving me nuts.
Please help.
EDIT: sorry, some images aren't uploading... reddit you suck.
