On Media Server Builds

I think I finally have a media server build that I'm going to stick with for a while. I have been going back and forth with different hardware configurations for a while now and it feels good to finally have a solution. For context, I run Jellyfin in a Docker container on a headless Debian system. I use mergerfs and SnapRAID for my storage pool as well as a couple of SSDs in a ZFS mirror all of which are accessible via Samba. My "media" folder lives under the mergerfs pool and is bind mounted into the Jellyfin container. For more info, see Alex Kretzschmar's excellent Perfect Media Server project.

When I first got the idea to build a media server, I asked my younger brother who had been into gaming for a while if he had any spare PC parts that he could sell me. He gave me an Intel Core i5 9600K and a B360M motherboard. After I started playing around with media server stuff on that machine, I realized that if I wanted to use Handbrake to re-encode Blu-ray rips, I was going to need something with a little bit more horsepower. At least that’s what I thought of the time because I imagined myself ripping 4K Blu-ray discs on the regular and I didn’t want every single one of those to take 12 hours to turn into a reasonably-sized version.

So I went on Facebook marketplace and was able to grab a system with a Ryzen 9 5900X, 64 GB of DDR4, and a B450 motherboard from ASRock for about 300 bucks. I soon realized that this motherboard didn’t have the expansion capabilities that I assumed were present given the number of slots. I also realized that I needed a discrete GPU to get any video output at all, which was kind of a pain and also upped the power draw so I got a Quadro P620 for 40 bucks or something like that on eBay. I also grabbed a 4U Rosewill server chassis that could hold up to 15 spinning hard drives for $100 on FB Marketplace. I figured this would be my all in one NAS/VM host/media server and I would use the Intel platform for testing and/or backups.

Very soon after this, I realized I could bring home some decommissioned hardware from my workplace -- I work as an Enterprise Applications programmer for an IT department. I wound up taking home two Dell OptiPlex 7090 small form factor desktop PCs, each with an Intel i7 10700 and 16 GB of DDR4. Since I had more hardware than I needed, I tried various configurations, including running Windows 11 on various machines to do the handbrake jobs. To cut a long story short, here is what I have learned and what I’m going to try going forward:

AMD makes great CPUs, but they’re just not the optimal choice for media server applications given their typical lack of integrated graphics. There’s also spotty support for hardware acceleration on the best free open source media server application Jellyfin and they have a higher idle power draw. Intel CPUs are the exact opposite with excellent hardware acceleration for live transcoding, excellent idle power draw and integrated graphics such that a discrete GPU is not required.

With that in mind, I will be running one of the i7s in a mini-ITX ASRock board that I got used for I think around $150. It has 2.5 gig networking, two M.2 slots and the x16 PCIe slot is able to be bifurcated should I need more expansion. This box will have 64GB of RAM, 8 cores, and over 40TB of useable storage. The 9th gen Intel platform will be my test system where I will try out new containers, play around with ZFS, learn about networking, and frequently wipe it clean.

The AMD platform with it's 12 cores and 64 GB of RAM will be serving as a workstation. I’m going to try running Ubuntu desktop but if I don’t like that, then I’ll probably run windows with WSL for development stuff. I think this will be nice as I can certainly blow through Handbrake jobs in half the time of the i7 chip. I’ll also be able to turn the thing off or let it go to sleep and conserve power.

One more thing I forgot to mention -- I also have a Beelink mini PC with an N100 CPU, 16GB of DDR5, and dual 2.5 gig ethernet ports. I originally bought three of these machines because I liked the idea of a low-power Kubernetes homelab but I quickly abandoned that idea after I realized how much needless complexity that would entail for my modest needs.

So now I’m running Proxmox as my base OS on my media server/NAS because I get ZFS support out of the box, a GUI for spinning up VMs/LXCs, and essentially bare-metal Debian where I run my Docker containers. I know, I know, you're not supposed to do that but I don't give a fuck -- it's a home server and I don't want to pass my iGPU through to a Debian VM and then lose graphical output on the host just so I can get hardware accelerated transcoding in Jellyfin. And I'm not adding a discrete GPU to the system either. "Umm why don't you just run Jellyfin in an LXC??" Um because fuck LXCs. I'm not parting ways with Docker Compose just so I can do things "right" by Proxmox's standards.

Oh and while we're on the subject, fuck running TrueNAS in a VM just so I can have Samba shares -- I already have a perfectly good Linux system with ZFS so why would I virtualize another entire Linux system running ZFS? Isolation? It's a home server. I don't need the extra security and I don't need five nines of uptime. I also just don't like TrueNAS because it won't let me install my own packages so I'm stuck with whatever their GUI has to offer. No thanks.

Of course, I might be wrong about all of the above. Only time will tell. As always, like comment, and subscribe, thanks for watching, and I'll see you in the next one.