Help with Plex server CPU decision

I am in the early stages of planning for a powerful (hopefully) Plex server and right now I am looking at different CPU options. I haven’t decided yet whether to build the server out of new parts or buy a second hand Dell or HP server on eBay and just add components as needed. I think I will do whatever gives the best results for the money.

The server will also be used for surveillance video and as a DAW server, so it needs some extra capacity. Plex would be the primary focus for this server though.

Right now I am trying to decide on what direction to take regarding CPU configuration for the server. It seems to be one of the most important things to decide early on since motherboard, chassis, RAM, Raid cards, etc depend on this decision.

What do you think of the CPUs below for Plex?

  • 2 x Intel Xeon E5-2667 v2 Q1 2014 - (3.30GHz, 8 cores), Passmark Dual CPU: 21600, Single Core 1941, Price 2xCPU: $300-$400 (eBay)
  • 4 x Intel Xeon E7-4890 v2 Q4 2015 - (2.8GHz, 15 cores), Passmark Quad CPU: 34472, Single Core 1780, Price 4xCPU: $400-$800 (eBay)
  • 1 x Intel i7 10900X Q4 2019 - (3.7GHz, 10 core) Passmark Multicore: 23264, Single Core: 2785, Price: $800
  • 1 x AMD Ryzen 3950X Q4 2019 - (3.5GHz, 16 cores) Passmark Multicore: 39390, Single Core: 2745 , Price: $700

I tried to include four CPU configurations that are quite different from each other and that also differ in age.

Would you recommend one over another and why?

If I go for a 2x or 4x CPU configuration I will probably start off with a second hand Dell or HP server. Is that a good idea? Are there any advantages/disadvantages to that compared to building my own from scratch? Possibility of adding well integrated separate storage chassis?

I primarily want to create a Plex server that is powerful and will last a few years. Cost will of course matter - but it will not be the only consideration. Power consumption and heat is not a big factor. Space used by the computer chassis is no big issue. I will probably rack mount the chassis.

Would a 4xCPU Dell server yield better results than the Ryzen or is it just silly to even consider it, because the Ryzen would be faster and cheaper anyway?

What would be the pros and cons of each of these CPU-configurations for Plex use? Is there a benchmark (or performance value) that would be usable for measuring Plex performance and approximately how well would they each do?

The 10900X is the only CPU with Quick Sync. Does this make a huge difference in transcoding performance? What if I add a GPU to the other CPU options for HW transcoding? Would that be possible and possibly even faster than a 10900X with Quick Sync?

Lots of questions. I know… Any ideas or thoughts to help me narrow down CPU alternatives would be greatly appreciated.

You spent a lot of time writing this extremely detailed post, and I commend you for that.

But I’m going to give a short reply, and I hope you don’t take offense to that.

Don’t combine everything you want into one box. It’s going to cost way more than it needs to.
Separate 2-3 machines out for dedicated uses:

  1. NAS
    Where all of your storage is located. The hardware requirements for a simple NAS are super low, like the NAS Killer 4.0.
  2. Plex (QuickSync)
    HP 290 or similar dedicated transcoding box, like the OTiS.
  3. BlueIris (QuickSync)
    Again, HP 290 or OTiS.

Even the lowest end recommended QS CPUs can handle 20+ transcodes or 6 1080p cameras.

Not offended at all. Thanks for your reply.

I had thought of separating tasks too and maybe I shouldn’t have mentioned that it might also do other things. I think separating things is the best way to go. I will definitely create a separate storage unit.

Regarding Plex, would you say Quick Sync is better than Xeon/Ryzen with GPU? What is the transcode performance of, lets say a Ryzen 3950X with a GeForce 2070 vs the 10900K?

Actually, I have to ask, is the Quick Sync transcode performance related to CPU performance at all or is it the same in all Intel CPU:s of the same gen? That is, does Quick Sync in an i9 10900K have the same transcode performance as Quick Sync in a Core i3 10100? In that case there doesn’t seem to be much point in buying a very high end Intel CPU for Plex?

The quad cpu box will probably cost you $100 a month just in electricity. Three separate boxes is the way to go in my book. I am currently downsizing from dual 2670 All in one server which are 8c/16t so 16c/32t total to a dedicated quick sync box just for plex (hp 290).

Yes the same generation intel Chip i3 vs i7/i9 has no difference is quick sync. Get the lower cost cpu for quick sync

I did some more research today and I thought I could share some links with info that I found.

Info regarding Quick Sync speed in different CPU’s:

Many of the the latest 9th and 10th gen Intel Core CPU’s that have a GPU seem to share the exact same UHD 630 GPU (in fact, maybe all of them do) and almost all of these UHD 630’s are more or less the exact same speed. There is just a 5-10% difference in UHD 630 turbo speeds between models. So, Quick Sync is the same speed in all the latest Intel CPU’s regardless of core count or CPU frequency, according to this list: https://en.wikichip.org/wiki/intel/uhd_graphics/630#Processors_with_UHD_Graphics_630

Info regarding Quick Sync vs dedicated GPU card:

I also found this video in my attempts to understand the speed of Quick Sync vs a dedicated GPU card: “Handbrake 4K H.265 video encoder speed test - NVENC v Quick Sync QSV - Intel i9 9900K vs RTX 2080 TI”: https://www.youtube.com/watch?v=1PqOInmAlIo

It seems to indicate about a threefold speed improvement going from Intel UHD 630 Quick Sync to nVidia GeForce 2080 for video encoding. I don’t know how this translates to Plex transcoding performance but I guess it has some relevance?

Actually, it is interesting that the difference is only 3 times considering the UHD 630 has a performance of only about 460 Gflops for single precision calculations (according to this: https://en.wikichip.org/wiki/intel/uhd_graphics/630#Performance) while the 2080 has a performance of about 9000 Gflops for single precision calculations, according to this: https://en.wikipedia.org/wiki/GeForce_20_series#Chipset_table

So, the 2080 has about 20 times more raw calculation power than the Intel UHD 630 and yet it can only encode video about three times faster. Either Quick Sync is very optimized for encoding (and maybe decoding too) or it skips on quality.

Maybe, if done right, a 2080 transcoder could actually be 20 times faster than UHD 630 Quick Sync?

My conclusion:

For Plex my conclusion right now is to do exactly what you guys recommended :slight_smile:

In the future when a dedicated GPU card has better drivers and performs 20 times faster than Quick Sync it might be a better option, but probably not right now.

I guess this means to use one of the latest Intel CPU’s with UHD 630 (Quick Sync) and select a CPU with just enough “passmark power” to handle the CPU related stuff it needs to deal with. From what I understand, it seems a 10th gen i5 or even i3 might be enough to create a pretty powerful Plex server.

I am a little bit disappointed to admit it, but the quad CPU purchase might have to wait :rofl: However, now I have to figure out how to build a NAS or DAS fast enough for all my needs.

Thanks for your reply.

It would have been fun to have a quad CPU computer though :smile: However, I can see it doesn’t seem to make much sense for Plex.

It seems like a huge jump in performance from the dual Xeon to a Celeron.

As far as I can tell the HP 290 with a Celeron G4900 has a UHD 610 GPU (https://www.intel.com/content/www/us/en/products/processors/celeron/g4900.html) and from what I could find out it has a performance of about 200 Gflops (https://en.wikichip.org/wiki/intel/hd_graphics_610#Performance).

Gen 10 Intel Core CPUs have a UHD 630 which can reach about 460 Gflops.

Do you know if the Quick Sync part of the GPU is in fact twice as fast in a UHD 630 vs a UHD 610 or if they are just about the same speed?

The compute performance of the graphics device is entirely unrelated to the encode/decode portion of the silicon - the differences in performance are according to the generational improvements of that silicon.

OK. Are the GPU cores not used at all for transcoding when Quick Sync does its “thing”? That is quite interesting. Makes you wonder what exactly Quick Sync is. That is, what exactly it is that it does (what types of computations) and how it does it.

Edit: Found this. It is old - but very interesting:

Regarding the difference in speed between quicksync and dedicated graphics card. What are we trying to get the plex transcoder to do? We are trying to get the video encoded/decoded for the end user to watch the video in realtime. As long as the transcoding can keep up with realtime playback, going any faster makes no difference.

If you had another use case like converting a big stack of videos from one format to another (what handbrake does) then the fastest speed for a given quality might be important as that big stack of videos might take 8 hours to convert on dedicated video card versus 24 hours on quicksync (as an example I just made up)