EPYC + Plex or Jellyfin or Emby

My only server play has been with Intel hardware, apart from a long time ago with 100-series Opterons. I am scheming on changing my setup, and I was wondering what EPYC CPUs can do compared to the Intel boys. I definitely am not interested in trashing either one -I was just wondering what your experiences were like. If you have used EPYC processors for such, what was the configuration? vCPU? How many cores? On your (home?) network? Rented? And so forth...

Thanks!

Comments

  • LabzeLabze Hosting Provider

    Depends on what you plan on doing with it. If you are just direct streaming then either is just fine. If you are talking dedicated/home servers then Intel have the benefit in the transcode department due to their iGPU and Quicksync. However, modern CPUs also handle transcoding quite well. I transcode 4k plex content on my 4 core 5950x VPS quite well. I know that is not EPYC but it is doable with enough cores.

    In the end, just look at the price/performance for your use case. Ignore whether is says AMD or Intel.

    Hostbrr.com - Cheap but powerful VPS - Instant setup
    AMD 7950XD and 5950X VPS starting at only $4.99! Order now!

  • tetechtetech OG
    edited May 2023

    These days I prefer to use a cheap GPU rather than rely on the processor. Recently started using Jellyfin with a few GPUs on a Proxmox VM. If you've got a good enough network you can optionally use rffmpeg to offload the GPU task to a remote machine. This more-or-less takes the CPU out of the picture for me, cost under $100 and uses only a few watts.

  • @tetech said: rffmpeg

    Thank you for the reply. I have never heard of rffmpeg before. I am not a GPU hound. I have 3dfx Interactive 3dfx Voodoo3 3000 (PCI) to Nvidia 2080 (PCI-Express), but I am not a gamer, and I do not pretend to be a GPU guru...from hence forth a GPuru. Your setup with Proxmox and a remote machine...are they on your local network? What are you transcoding? Which GPUs? I am interested. It is hot as dicks on fire where I live, so I am keen to understand how many watts I am adding to our local situation.

    @Labze said: I know that is not EPYC but it is doable with enough cores.

    How many cores to transcode 1080p ~ @22Mbps -> 1080p @3MBps?

  • @hornet said:

    @tetech said: rffmpeg

    Thank you for the reply. I have never heard of rffmpeg before. I am not a GPU hound. I have 3dfx Interactive 3dfx Voodoo3 3000 (PCI) to Nvidia 2080 (PCI-Express), but I am not a gamer, and I do not pretend to be a GPU guru...from hence forth a GPuru. Your setup with Proxmox and a remote machine...are they on your local network? What are you transcoding? Which GPUs? I am interested. It is hot as dicks on fire where I live, so I am keen to understand how many watts I am adding to our local situation.

    I could not be even remotely considered a gamer. I get the GPU solely for video processing (and recently I'm starting to use it for motion detection on surveillance cameras, completely different story). My preference is for nvenc (Nvidia cards ideally with a Turing engine or better) because these have most hardware encoders people need. For example, I picked up T400's for under $100 on eBay, and these fit in a single x16 PCIe slot without the need for additional power, which makes them good for a server.

    Yes, this is on a home network. Very roughly, old (E5-2630 v3) CPUs, Proxmox, create debian VM, pass through T400 GPU to it, install Nvidia drivers, install Jellyfin, configure it to use nvenc, and then I use a HDHomeRun for a TV tuner. Wireguard to HAProxy on an Oracle instance to avoid disclosing home IP. Transcoding is either live TV or the video library to match a target bitrate of the browser/player. In very, very informal tests transcoding live TV added around 5-10 watts max, it also used one vCPU so not all of that is from the GPU.

    I don't need rffmpeg in that situation. Just saying that (depending on network) GPU and video server don't necessarily even need to be on the same server. It is really designed for Jellyfin to farm stuff out for transcoding on a "GPU array" but I use it the other way around... taking video on a different VM and transcoding it on the GPU tied to the Jellyfin VM.

  • Would def be leaning strongly towards intel for quicksync on any build vaguely in the plex/emby/jelly direction

    @tetech said: I get the GPU solely for video processing

    I gather intel Arc cards are gaining popularity for this too

Sign In or Register to comment.