A more performant Plex Media Server build

Hardware Hub Build a PC
1

I’m currently using on my QNAP NAS systems as the PMS, but unfortunately, because it has an Annapurina Labs quad-core ARM processor in it, scanning the media (or doing anything else with it) is quite slow.

My thought right now is that because I LIKE the fact that typically, the NAS units consume less power than an actual server, my thought is that I would just get a mini PC (whether it’s from Minisforum, Beelink, Morefine, ASRock Industrial, Intel NUC, etc.) and use the NAS to be the “dumb” data server, but then use the mini PC to actually do all of the “heavy lifting”/data processing work.

I don’t know what would be the implications of trying to, for example, run the media scan over a gigabit ethernet network. (Some of the mini PCs have a 2.5 GbE NIC, but I don’t have any 2.5 GbE capable switches at the moment.)

Does the community here think that this would work or would this be an absolutely terrible idea?

What would be the catches/gotchas that I should pay attention to/be aware of if I am going to try and set up something like this?

Your thoughts would be greatly appreciated.

2

Media scanning over 1gbe would be perfectly fine. If you want something cheap and low power, find a small form factor pc with a recent Intel CPU and just use Plex with QuickSync. Find one you can put an nvme drive in so your thumbnails, cache, etc are quick.

3
4

How tied are you to Plex?

I’ve been using Kodi (coreelec specifically), to play mounted media, … and there’s also Jellyfin if you’re looking for a more mounted solution.

5

Yeah, I tried that.

QuickSync is useful for transcoding video, but not audio.

(i.e. on my Samsung TV, it can direct stream/play the 2160p video, but it has to transcode the audio down from DTS:X to PCM, and QuickSync can’t help with that.)

@jode
Yeah, I saw that video.

I found a chassis (SILVERSTONE Black Grandia Series SST-GD08B ATX Media Center / HTPC Case - Newegg.com) that would work well for something like this, where the chassis is not quite a full rackmount size, (something a little bit smaller), but is also able to accept 8 3.5" HDDs.)

(I’d still have to resolve the user permissions conflict issue that I am having between NFS and SMB, but that’s a different problem altogether.)

I am only “tied” to it because per Linus (he’s not changing from Plex for his home theatre room), because for what he needs it to do, it works fine.

(local playback/streaming)

And for my wife and kids, I’d have to agree with that.

I did try using one of my other mini PCs which has the AMD Ryzen 9 5900HX (8 core/16 thread)/64 GB RAM system to do the intro detection over the network, and it was pegging the GbE NIC at almost 100%, reading the media files over the network (do I “need” 2.5+ GbE?).

But also interestingly enough, i’m not sure if there is a bottleneck somewhere (whether it’s from the 8 HGST HDDs) or it’s because I’m using GbE or if it is because Plex’s intro detection algorithm can’t process multiple video files simultaneously, and therefore; can’t fully load up my 8-core/16-thread 5900HX CPU.

Therefore; given these questions (about the bottleneck), is what my NAS doing, already really doing it about as fast as it can go, and that this project proposal won’t really result in a significant improvement in intro detection speed, because of any of the combination of the aforementioned potential bottlenecks?

(I did think about putting the system all together, in the same chassis, so that it’s not having to read the media files over ANY network at all, but when running off of eight HDDs, I’m also not 100% convinced that that would matter that much nor it would make that much of a difference, because it is reading the media files off of 8 HDDs.)

And I try not to use SSDs as I have a tendency to burn through the write endurance limits. (I’ve killed 6 SSDs in the past 5 years already, so I have, effectively decommissioned almost all of my SSDs.)

6

The key takeaways for me were that the modern CPUs are pretty power efficient at idle, the bulk of power consumption coming from feature rick mobo components. The main power savings stem from using mATX boards come feature resticted (relative to full sized boards) and build with power savings in mind.
I just “upgraded” my 24/7 server from an ASUS WS X570 ASE mobo to a ASUS TUF GAMING B550M-PLUS WIFI II saving 40W at idle in the process.

I am using this chassis. If you get the rackmount ears for it it resembles a short 4U chassis.
A full sized ATX board will reach all the way to the drive bay making space tight.
I use a Ryzen 5900X with a beQuiet DARK ROCK TF2 cooler that fits nicely, but the case so shallow that it may be more suited for a less crowded mobo that I have. You may have picked up that this is not my power saving rig.

7

There is no hardware transcoding for audio in Plex, regardless. That limitation is not exclusive to QuickSync.

8

I understand that.

However, power efficiency really isn’t (much) of an objective for this post.

I think that my eight HGST 10 TB SATA 6 Gbps HDDs will probably consume more power than the rest of the system anyways.

And that’s also always the optimisation problem - is to balance or optimise between power efficiency (whether it’s at idle or at full load or single-threaded full load) vs. cost vs. overall performance.

ARM processors can be very power efficient, but they’re also REALLY, REALLY slow at doing computationally intensive tasks, that when you integrate power with respect to time, the total amount of energy that you consume might end up being higher than if you just picked something faster, but also consumes more power, but is able to finish the task sooner, and therefore; consumes less energy overall.

Unfortunately, there really isn’t a good database for single threaded power consumption as most tend to measure ONLY at idle and ONLY at full load (typically all cores at 100% or as close to that as possible) vs. only loading one core/thread up to 100%.

The case that the aforementioned video uses as the thumbnail - I actually have that Supermicro case already (that’s currently used for my micro HPC cluster headnode).

So, in theory, I can always migrate that system out of that case and into another case, and then use that exact same case for this system instead, and that way, I would be able to use the 8 3.5" hot swap drive bays along with the SATA 6 Gbps backplane that I had swapped out with a SAS 12 Gbps backplane (for my micro HPC cluster headnode).

I haven’t quite fully decided on what I want to do with my server(s) just yet because there was also another proposal that was introduced where I moved all of the hard drives from my various NAS systems into a single 36-bay chassis, and then drive the system entirely off a single processor (haven’t decided which one yet, but am leaning more towards AMD Threadripper because of PCIe connectivity demands.)

(i.e. the SAS RAID HBAs will need I think two PCIe 3.0 x8 slots, and then if I add a quad 2.5 GbE port NIC to it, that will consume I think another PCIe 3.0 x4 slot (or maybe it’s a x2 slot - hard to tell from the pictures), plus a quad 10 GbE SFP+ card (which are very expensive right now), which will likely consume probably at least a PCIe 3.0 x8 slot, plus my dual 100 Gbps IB NIC, which requires a full PCIe 3.0 x16 slot. So that’s a total of 44 PCIe 3.0 lanes.

I can probably feed at least that many lanes with an Asus P9X79-E WS motherboard and maybe like an Intel Core i7 4930K as well, but that will only support 40 PCIe lanes, rather than the 44 lanes that the current count requires (not including any other peripherals on the motherboard that will also require PCIe lanes as well).

But that probably WON’T be a power efficient system neither, due in part to this total PCIe bandwidth requirement.

9

Oh, I know.

I was only specifically addressing the point in regards to QuickSync being explicitly mentioned.

I tried it with NVENC as well, which of course, didn’t work, which points me in the direction of trying to find a really fast processor for single threaded application (assuming that I don’t consolidate my servers into a single system).

10

gotcha, i understand. for what it’s worth, i had no issues with multiple streams transcoding audio on just a Celeron G5905. the only reason i upgraded from that to the current i3-10105 is because i added Frigate to that machine.

11

i am running 2x perc h330 SAS cards in PCIe 2.0 x8 slots in an ancient opteron 6366 with a 10g nic in a PCIe 2.0 x4 slot. this NAS is storage for a huge EMBY server, nextcloud, about 5 mapped drives, and a proxmox server connects to it also.

all i have for storage is 6x 12tb sas helium drives in mirrored pairs, 3 vdevs in a zpool. and then 4x sata 480gb ssd drives in a separate pool.

the emby server is in proxmox and accesses the data across the 10gb nic. ive got a house full of people all day long and other various family and friends and this stuff runs all the time. i’ve not ever had performance issues with it.

my point here is PCIe is not a bottle neck. you will cap out something else well before, (unless you have 16x sas ssd in there).

12

Not sure about those specifically, modern drives use 5-10W typically except during initial spin up… so 50W - 100W? … (Probably ballpark $50-200 per year depending on where you live).

13

I think that it might make a difference depending on what you are transcoding the audio to/from.

I’ve found that even with a Ryzen 9 5950X, because the audio transcoding isn’t multi-core/multi-processor capable, so even that struggled to keep up with transcoding the audio from DTS:X 7.1 down to PCM or AAC.

(The playback on my Samsung TV keeps stuttering because the server can’t keep up with the audio transcode, even when I’ve set up a separate Plex server on said 5950X system, JUST to test the audio transcoding out.)

I COULD optimise the media though, and then that would play better, but I really don’t want to have two copies of every single thing that’s in my Plex library (the original plus the optimised copy).

Not in terms of bandwidth, but it is a bottleneck in terms of the number of PCIe lanes that a CPU will supply.

(i.e. if all of the add in cards take up 44 PCIe lanes and say, I’m only using a Ryzen or the latest Core i7/Core i9 processor – those CPUs will only supply upto 24 PCIe lanes at best. As a result of that, typically, the motherboard will run out of the appropriately sized PCIe slots regardless of whether those add in cards will actually consume and use all of the bandwidth that’s available to them.)

I’ve yet to see a X570 or Z690/Z790 motherboard that has like 7 full size PCIe x16 slots (even if they aren’t all electrically wired for x16 operation).

Conversely, sTRX4 and sWRX8 motherboards WILL have at least the PHYSICAL slots be x16, again, even if not all of them are wired electrically for x16 operation.

Yeah, I think that spin up is ~30 W mark.

Idle might be down to about 5 W, but active use, it’s probably higher.

(All I can tell is that my 12-drive, dual Xeon server that it’s sitting in right now, idles at around 250 W, which is a LOT higher than my QNAP NAS units.)

I was looking at replacing all of the systems and consolidating them down to a single 4U, 36-bay chassis, and the power supplies on those are redundant 1280W power supplies.

Therefore; I would have reason to believe that a part of the reason why the power supplies have such a high power rating is because it is expected or anticipated that the rest of the system itself might also be quite power hungry.

According to my UPS right now, all of the servers combined suck back around 650 W at idle. (Two QNAP TS-832X, one QNAP TS-453 (I think), plus my 12x HGST 6 TB dual Xeon E5310 TrueNAS (which has an 80 W TDP).)

So, if the power supplies are going to be rated that high, it would be a bit of a pointless endeavour, trying to consolidate those systems down to one, and then ending up where the power consumption would be identical or higher than what it was before.

14

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.