omv6 and nvidia gpu

Zitat von Cap10Canuck

Tried to install the Nvidia drivers again as per the install guide and the install stopped at the following:

"This system has a graphics card which is no longer handled by the NVIDIA driver (package nvidia-driver).

You may wish to keep the package installed - for instance to drive some other card - but the card with the following chipset won't be usable.

06:00.0 VGA compatible controller [0300]: NVIDIA Corporation GK104 [GeForce GTX 760] [10de:1187] (rev a1)

The above card requires either the non-free legacy NVIDIA driver (package nvidia-tesla-470-driver) or the free Nouveau driver (package xserver-xorg-video-nouveau). Use the update-glx command to switch between different installed drivers. │

Before the Nouveau driver can be used you must remove NVIDIA configuration from xorg.conf (and xorg.conf.d/)."

At this point I may just throw in the towel, given it errored-out at only step 3 of a relatively complex set of installation instructions.

Alles anzeigen

Cap10Canuck:

If you really want to use such an old graphics card (in my opinion there won’t be a big benefit) you can follow this instructions to install the correct nvidia driver:

NvidiaGraphicsDrivers - Debian Wiki

Then proceed the tutorial after the nvidia driver part. But I‘ve never tried using the legacy drivers.

For me it would make more sense to get a used nvidia quadro p600 or p1000 wich is consuming way less power and has way more video performance.

ok, i spoke to soon, drivers are installed, GPU should be visible inside docker containers as confirmed by

docker run --rm --runtime=nvidia --gpus all nvidia/cuda:11.6.2-base-ubuntu20.04 nvidia-smi

i have changed plex and immich settings to utilize HW accel, but it does not work. neither plex or immich utilize the HW acceleration... what else could have i missed?

Zitat von molnart

ok, i spoke to soon, drivers are installed, GPU should be visible inside docker containers as confirmed by

Code

docker run --rm --runtime=nvidia --gpus all nvidia/cuda:11.6.2-base-ubuntu20.04 nvidia-smi

i have changed plex and immich settings to utilize HW accel, but it does not work. neither plex or immich utilize the HW acceleration... what else could have i missed?

Hard to say without any information about your installation. In general it is only possible to use one GPU per container. You can not use the same GPU in multiple containers.

HW transcoding will only be used, if the codec of the video can be processed by your GPU. Wich graphics card are you using ?

Zitat von chris_kmn

You can not use the same GPU in multiple containers.

wtf, you serious? thats a serious bummer.

after further testing is looks like always the container last spun up has access to the GPU, the others don't. so far I can occasionally get HW acceleration for Plex and Ollama, but its pretty unpredictable which one can actually use it. and additionally i wanted to play around with fooocus image generation...

so i guess my only option is to to move my GPU apps to separate LXC containers, as there seems to be a way how to share the GPU between multiple LXCs

Zitat von molnart

wtf, you serious? thats a serious bummer.

after further testing is looks like always the container last spun up has access to the GPU, the others don't. so far I can occasionally get HW acceleration for Plex and Ollama, but its pretty unpredictable which one can actually use it. and additionally i wanted to play around with fooocus image generation...

so i guess my only option is to to move my GPU apps to separate LXC containers, as there seems to be a way how to share the GPU between multiple LXCs

I‘m not 100% sure, but I con only run one container with one GPU and I‘ve read about that somewhere. So it seems to be valid.

May be there are ways to overcome that issue, but I don‘t know them. Strange thing on my machine is, that Plex ist running together with Tdarr, using the same GPU. But Plex and Handbrake are not able to use the same GPU. So I‘m running Handbrake on the internal GPU (Intel) and Plex on nvidia.

It might be connected to the nvidia container toolkit. It passes the nvidia driver, which is outside of the container, to the docker container. There might be ways to install the nvidia driver inside the container. But I‘ve never done that.

Zitat von chris_kmn

For me it would make more sense to get a used nvidia quadro p600 or p1000 wich is consuming way less power and has way more video performance.

Thanks - that is now the plan! I am at the bottom of the spare parts box, so will be looking for an old GTX 1660, which still has a bit of oomph.

What would be a good GPU for transcoding looking to upgrade from a GeForce GTX 1050

I would recommend a nvidia quadro p600-p200 or a quadro t600. They are using less power and are pascal/turing gpu‘s.

You can also check this matrix, wich gpu‘s support what codecs: navida codec matrix

it seems after a kernel update my nvidia gpu is just "gone". all my containers using nvidia refuse to start, nvtop or nvidia-smi shows there is no gpu in the system. lspci shows it.

Which kernel are u using now?

Zitat von molnart

it seems after a kernel update my nvidia gpu is just "gone". all my containers using nvidia refuse to start, nvtop or nvidia-smi shows there is no gpu in the system. lspci shows it.

kernel updates often require nvidia driver reinstalls because the install builds a kernel module against the kernel version. A new kernel version means the existing module is no longer valid.

how do i do that? i tried dkms autoinstall, update-initramfs -u, update-grub, even apt install --reinstall nvidia-driver but nothing seems to work. if i need to purge the drivers or even worse restart OMV i rather throw that card out from a running system directly into the trash, but there is no way I am restarting my system before it reaches 500 days uptime.

Zitat von molnart

but there is no way I am restarting my system before it reaches 500 days uptime.

You can't kernel upgrades if you don't reboot.

There's no prize for "biggest update time",

Zitat von Tallboi

What would be a good GPU for transcoding looking to upgrade from a GeForce GTX 1050

Personally I am using a Quadro p2200 very powerful and not too expensive low consumption

About kernel update on my system with a Nvidia direct install I just do

cd /srv/dev-disk-by-uuid-1942e416-bbf4-4f33-b6a3-00171f82bf75/
sh ./NVIDIA-Linux-x86_64-550.67.run

And reboot, take 2mn.

turns out the key to rebuilding the modules is the following:

sudo dpkg-reconfigure nvidia-kernel-dkms
sudo nvidia-ctk runtime configure --runtime=docker
sudo systemctl restart docker