Add a discrete GPU or not?
Add a discrete GPU or not?
In my server I currently have an Intel i7 9th gen CPU with integrated Intel video.
I don't use or need A.I. or LLM stuff, but we use jellyfin extensively in the family.
So far jellyfin worked always perfectly fine, but I could add (for free) an NVIDIA 2060 or a 1060. Would it be worth it?
And as power consumption, will the increase be noticeable? Should I do it or pass?
QuickSync is usually plenty to transcode. You will get more performance with a dedicated GPU, but the power consumption will increase massively.
Nvidia also has a limit how many streams can be transcoded at the same time. There are driver hacks to circumvent that.
Most Intel GPUs are great at transcoding. Reliable, widely supported and quite a bit of transcoding power for very little electrical power.
I think the main thing I would check is what formats are supported. If the other GPU can support newer formats like AV1 it may be worth it (if you want to store your videos in these more efficient formats or you have clients who can consume these formats and will appreciate the reduced bandwidth).
But overall I would say if you aren't having any problems no need to bother. The onboard graphics are simple and efficient.
If it is working for you as is, no need to make a change
For an old nvidia it might be too much energy drain.
I was also using the integrated intel for video re-encodes and I got an Arc310 for 80 bucks which is the cheapest you will get a new card with AV1 support.
Intel has excellent transcode, even in their igpus.
I use an arc750 specifically for transcode, av1 runs at ludicrous speeds, but don't do an Nvidia, they kind of suck because they dont support vaapi, only nvenc/nvdec and vdpau.
The Intel GPU is probably better
Look up the GPU on these charts to find out what codecs it will support: https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new
NVENC support will tell you what codecs your GPU can generate for client devices, and NVDEC support determines the codecs your GPU can read.
Then compare it with the list of codecs that your Intel can handle natively.
Thanks!
Both the 2060 and the 1060 don't support AV1 either way, so I guess its pointless to me.
If the iGPU is getting the job done, I would leave that alone. You could add a GPU and pass it through to a gaming VM. But that is an entirely different project.
Could be an interesting project tough, will definitely think about that. Not top priority, but why not since the hardware its free?
I would avoid it, if you care at all about availability and downtime. The result will probably not be great, you need to ensure the server side gets enough resources under load, and setting it up may require constant restarts if things aren’t immediately working as expected.
Nonetheless, here is a link where someone did essentially exactly that on NixOS: https://astrid.tech/2022/09/22/0/nixos-gpu-vfio/
I ran a 1650 super for a while. At idle it added about 10W and would draw 30-40W while transcoding. I ended up taking it out because the increased power wasn't worth the slight performance increase for me.
Yeah look like a lot... Probably not worth it.
I only have a GPU because my CPU doesn't have any graphics. I don't use the graphics anyway, but I need it to boot. So I put our crappiest spare GPU in (GTX 750 Ti) and call it good.
I wouldn't bother. If you end up needing it, it'll take like 15 min to get it installed and drivers set up and everything. No need to bother until you actually need it.
Host steam-headless and use the GPU for that so you can have remote gaming on your phone anywhere you have 5G
Holy shit that’s awesome.
My work gives me unlimited uptime on a dedicated A10. AI engineer working for NVIDIA. Doubt they’d care if I set something up like this, or if they’d even know.
But I have a 4090 at home anyway so like, do I even need or is this just another way to explore infinite hobby of tinkering with computing.
Cool idea!