#Why is my gpu not detected?

1 messages · Page 1 of 1 (latest)

covert moon
#

I have an old gaming laptop that I've set up as my HA server, it has 32 GB ram, a 9750h, and an rtx 2060 mobile. In the hardware tab I can detect the ram and the processor fine, but I don't see any graphs for the gpu itself, which I want to use specifically for the voice assistant processing. Is it detected but I'm not seeing it or do I have to somehow enable it?

jade jolt
covert moon
#

yeah the x86/x64

jade jolt
# covert moon yeah the x86/x64

HAOS itself does not have drivers to use the GPU.
you are better off installing proxmox, and HAOS on a VM. that way you can install gpu drivers on proxmox and run the stuff you need to use it within LXC's

covert moon
#

wait so is proxmox the main OS?

jade jolt
#

proxmox is a hypervisor

#

but yes you would install that "on the metal"

#

you just wanting to run whisper with gpu? or also a small llm too? how much vram are you working with?

covert moon
#

6gb

#

with a small llm

#

i know it's not all powerful but the laptop isn't really doing anything

jade jolt
#

yeah you could probably run whisper and something like a 3B model

#

maybe a bit limited but would be cool in any case

#

have you ever used proxmox or any other hypervisor before?

covert moon
#

i've never heard of it before

jade jolt
#

so its base OS that runs virtualised systems, either as VM's or LXC containers. it is itself debian linux under the hood

#

when it comes to proxmox setup, there is a bunch of "community scripts" which help a lot

so this will be a brief breakdown and you may need to research the steps (happy to answer some questions too) but the basics of what i would suggest is:
download and install proxmox ve -ISO HERE
once its installed on its webui you can access the shell (or use it on the system or ssh)
run the POST INSTALL script
then you can run the HAOS INSTALL script which sets up a HA VM for you
then you can run the DOCKER LXC script, this will set up a LXC container with docker installed.

then you will need to ensure that nvidia drivers are installed on proxmox, then on the LXC, and the nvidia docker stuff too.
then you can install a whisper(with gpu support) docker container on the LXC
then also install ollama docker container on the LXC

then you connect whisper and ollama to HA and your "good"

covert moon
#
  1. direct passthrough 1physical gpu to 1vm?
  2. use gpu on pve host?
  3. separate gpu to vgpu and then pass to vm?
#

which one of these options do I want?

jade jolt
#

you have the GPU on the host, which can then be accessed and used by LXC's

#

so... "2" i guess

#

so you currently have proxmox ve installed with a haos vm and a docker lxc running?

covert moon
#

yes

#

i did accidentally give haos 4 cores and dockler 2 cores for the cpu is that ok?

#

my cpu is 6 cores overall

jade jolt
#

thats fine. and you can change this down the line if you need to

#

whilst at this point we should check your haos vm is using "host" cpu type too

covert moon
#

yeah it says host

jade jolt
#

ensure it says "host" here

#

cool

#

right so now you need to ensure nvidia drivers are installed on proxmox

covert moon
#

when I download the driver from the website should i use the linux 64 bit opr the linux aarch64 driver?

jade jolt
#

you should be able to installl it with apt

covert moon
#

i don't know what that is

jade jolt
#

apt is the package manager for debian and debian based systems

covert moon
#

sorry im very new to proxmox

jade jolt
#

i havent actually installed nvidia drivers on proxmox before. let me check the repos

#

ok so maybe you cant do it with apt. you will need to get some stuff with apy and also indeed download from the website

covert moon
#

I found this link idk if its the right approach

jade jolt
#

yes, this guide actually looks pretty decent

jade jolt
#

hopefully you can work your way through that and get to the point where nvidia-smi runs and sees the gpu

covert moon
#

this doesn't show up in the guide, which one should I go for?

image.png
jade jolt
#

nvidia properatory i think

#

they are more functional than the open stuff currently

covert moon
#

what does this mean?

image.png
jade jolt
#

ah thats the opensource driver

#

on the proxmox terminal run
echo "blacklist nouveau" >> /etc/modprobe.d/blacklist.conf
then reboot the system

#

then try again

#

that should block the nouveau driver from running

charred pelican
#

The nvidia installer should do this itself.

jade jolt
charred pelican
jade jolt
charred pelican
#

I think it didn't happen to me because the GPU isn't used as video output but I know that the driver package ships with a blacklist file.

covert moon
#

ok i think it's installed but next to the gpu name it just says off is that normal?

image.png
jade jolt
#

yeah i think so

covert moon
#

ok so the next step is downloading the drivers on docker too?

#

wait I just realized my bluetooth card is no longer recognized

jade jolt
#

do you mean on haos? you will need to pass it to the vm for haos to see it

#

can help you with that but, one issue at a time 😛

covert moon
#

ok lets do the gpu one first

#

switchbot temperature sensor can wait

jade jolt
#

so we want to ensure the LXC has the gpu as a "resource" and then ensure drivers are installed on the LXC and also you will need to install the nvidia container toolkit

covert moon
#

so add it as a device passthrough?

jade jolt
#

yeah on the lxc resources tab you add as device passthrough and then need to put in the device paath

covert moon
#

how would I find the device path?

jade jolt
#

based on impacts guide

#

you can add them on the gui too. i am not 100% how it works when needing the gid (which you can set in advanced)

#

should work out though. you have the paths there and if it says to include a gid you can add it in the advanced tick