41
Hardware Acceleration in Linux in Proxmox
(lemmy.world)
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
this post was submitted on 06 Mar 2024
41 points (97.7% liked)
Selfhosted
40133 readers
525 users here now
A place to share alternatives to popular online services that can be self-hosted without giving up privacy or locking you into a service you don't control.
Rules:
-
Be civil: we're here to support and learn from one another. Insults won't be tolerated. Flame wars are frowned upon.
-
No spam posting.
-
Posts have to be centered around self-hosting. There are other communities for discussing hardware or home computing. If it's not obvious why your post topic revolves around selfhosting, please include details to make it clear.
-
Don't duplicate the full text of your blog or github here. Just post the link for folks to click.
-
Submission headline should match the article title (don’t cherry-pick information from the title to fit your agenda).
-
No trolling.
Resources:
- selfh.st Newsletter and index of selfhosted software and apps
- awesome-selfhosted software
- awesome-sysadmin resources
- Self-Hosted Podcast from Jupiter Broadcasting
Any issues on the community? Report it using the report flag.
Questions? DM the mods!
founded 1 year ago
MODERATORS
Don't do that. Run Jellyfin in its own VM with a GPU passed though via pcie passthough.
That's going to be almost impossible to do with an iGPU. Makes way more sense to pass through to LXC.
It takes about 2-3 clicks. What do you mean impossible? LXC is likely faster but it takes more setup.
2-3 clicks? That's hilarious!
These are the steps it actually takes: https://3os.org/infrastructure/proxmox/gpu-passthrough/igpu-passthrough-to-vm/
That's the best case scenario where it actually works without significant issues, which I am told is rarely the case with iGPUs.
In my case it was considerably more complicated as I have two GPUs from Nvidia (one used for host display outout), so I needed to block specific IDs rather than whole kernel modules.
Plus you lose display access to the Proxmox server which is important if anything goes wrong. You can also only passthrough to one VM at a time. Compared to using LXC you can passthrough to almost unlimited containers, and still have display output for the host system. It almost never makes sense to use PCIe passthrough on an iGPU.
The reason to do passthrough is for gaming on Windows VMs. Another reason is because Nvidia support on Proxmox is poor.
This is a guide to do passthrough with LXC: https://blog.kye.dev/proxmox-gpu-passthrough
It's actually a bit less complicated for privileged LXC, as they are having to work around the restrictions of unprivileged LXC containers.
Its always worked well for me. I passthough my dedicated graphics and USB controller to a Pop os VM and then the integrated graphics to the Jellyfin VM. I initially had to enable virtualization extensions and for the dedicated graphics there was a bit more setup but for the most part it is reasonable.
My point is it's not actually much (or potentially any) simpler to use PCIe passthrough than using an LXC. Yet it comes with more resource usage and more restrictions. Some hardware is more difficult to pass through, especially with iGPUs. I don't even think all iGPUs even use PCIe.
iGPUs are incredibly easy to pass though and are PCIe devices.
Not all of them. Have a look at a Raspberry Pi or Apple Silicon devices. In fact most ARM SoCs I am fairly sure don't use PCIe for their iGPUs. This makes sense when you think about the unified memory architecture some of these devices use. Just in case you aren't aware Proxmox does indeed run on a raspberry pi, and I am sure they will gain support for more ARM devices in the future. Though I believe an x86 device with unified memory could also have problems here.
If it wasn't connected via PCIe how would it talk to the GPU. Anyway Proxmox does in fact not officially support ARM so that is a pretty miniscule use case. I'm not even sure why you would want Proxmox on a low powered device.
For me PCIe pass though is the easiest. Virtualization adds little overhead in terms of raw performance so it isn't a big deal. If you prefer LXC that's fine but my initial statement was based on my own experiences.
AMBA/AXI-bus in the case of the Pi. GPUs existed long before PCIe did lol.
One some x86 systems the CPU and GPU aren't connected with PCIe either. AMD has infinity fabric that they use for things like the Instinct MI300 and some of their other APUs
Edit: Oh yeah also ARM isn't just low power anymore. It's used in data centers and super computers these days. Even if it was there is lots of stuff you can do with a low power node, including running file servers, DNS or Pi hole, web servers, torrent/usenet downloaders, image and music servers, etc. I have also seen them used to maintain cluster quorum after loss of one more powerful node. A two node cluster won't have quorum if one fails, so adding a pi as a third node makes sense.