search

DualCoder / vgpu_unlock

Unlock vGPU functionality for consumer grade GPUs.
MIT License
4.45k stars 422 forks source link

Single instance on 2080ti framebuffer 11GB #112

Open efschu opened 1 year ago

efschu commented 1 year ago

What would be the value for a modified profile for a 2080ti which makes it use all of the available VRAM (11GB) for one instance ?

When 1GB is 0x3B000000 should 11GB be 0x289000000, correct?

bossrhino commented 1 year ago

Can I ask a question? How well is the 2080ti working for you? I'd like to purchase either a 2060 12gb for VDI + Plex/Jellyfin + AI Development Server + Docker Hardware Passthrough.

Not interested in raw horsepower, but want plenty of memory for applications.

efschu commented 1 year ago

It works fine so far, but didn't figure out how to give a single instance the whole memory.

So for now, I "switched" back to simple passthrough.

hetzbh commented 6 months ago

IMHO, if you're going to give the whole memory, you should do a GPU passthrough, and the performance would be greater (since there's no overhead to deal with)

efschu commented 6 months ago

Actual the "overhead" of VGPU compared to normal passthrough is minimal. Running 3Dmark I get same score.

But you cannot use all GPU memory in VGPU setup.