I’ve been testing my PMS on a SuperMicro/Xeon with a PassMark of ~17500. Obviously, it only does software transcoding. I realized my SuperMicro/Xeon can barely handle 4K HDR (Blu-ray Remux, 10-bit HEVC) to 10Mbps 1080p (streaming). I’d like to be able to have the CPU to handle subtitles and 4K HDR (Blu-ray Remux, 10-bit HEVC) to 40 Mbps 4K (streaming).
As such, I want to build a dedicated Plex server.
Form Factor: “Quiet” Mini-ITX
CPU: Intel i9-13900 (13th Gen, Raptor Lake, UHD 770)
Storage: Already have a Synology NAS with a ton of 4K HDR Blu-ray Remux’s
Network: Ubiquiti 1G Network
ISP: 1G up/down
Goals:
4K HDR (Blu-ray Remux, 10-bit HEVC) to 10Mbps 1080p (streaming) for the gym (iPad)
4K HDR (Blu-ray Remux, 10-bit HEVC) to 40Mbps 4K (streaming) for travel (Apple TV)
I like subtitles!
I’m a little confused about the “Intel Xe Graphics Architecture (Gen 12)” on my ASRock Z790M-ITX.
Questions:
What’s the correlation between my motherboard’s Intel Xe and my CPU’s i9-13900’s UHD 770 with QuickSync? How I’m understanding it is the motherboard has an integrated Intel Xe GPU? And the CPU has integrated UHD 770 GPU?
Can PMS take advantage of Intel Xe? I don’t think so, right? PMS can only use QuickSync and NVENC.
Do I need to tell PMS to not sure the motherboard’s Intel Xe and use the CPU’s QuickSync?
Basically, WTH does “Intel Xe Graphics Architecture (Gen 12)” mean?
To be honest, TBD on how many simultaneous streams that I want to transcode. If this project works well, I’m sure friends and family would want to use it. And if I’m going to build a dedicated Plex server, I don’t mind ponying up for an i9. It’s less headache than upgrading later.
I won’t be putting much money into storage/memory on this dedicated i9 Plex server as it’s not using much storage/memory on my SuperMicro/Xeon.
Intel XE is an architecture that optimizes how UHD 770 is delivered from the CPU and, in this case, is used as a marketing term by Asrock to explain how the CPU integrates with the onboard HDMI/DP to provide video without a dedicated card. Intel does have an iGPU line called the Intel Iris XE which leverages the architecture, and does work with Quick Sync, but you’ll really only see those in laptops and NUCs.
With that said, if you’re looking for 4k transcoding, UHD 770 is probably your best bet for quality – Not many people recommend transcoding 4k, but the UHD 770 is a monster. How many streams you’d get out of it is unknown to me.
It’s new hardware and not a lot of people have moved to it. I have a separate thread open investigating some abnormal quality issues with UHD 770 and Intel Iris XE with subtitles, so it might be worth checking that out also.
I think you might even be surprised how much adding a quadro P2000/P4000 might help your existing system with the Xeons. (if pcie is an option).
Honestly, overkill is probably an understatement. If you’re going to be doing transcoding, the UHD is going to do most all of the heavy lifting. You can find the UHD 770 in 12th gen i7s to save change. If you’re really set on the i9 13th, you might consider leveraging that system for other things (not just plex) – eg. virtualization host.
Got it. “Intel Xe Graphics Architecture (Gen 12)” as ASRock’s marketing fluff makes a lot of sense. I was Googling the crap out of “Intel Xe, Plex” was getting confused. I even read the ASRock manuals, schematics, and BIOS settings to try to make sense of it, haha.
Yeah, I wanted UHD 770 for quality with a lot of extra horsepower for subtitles and audio transcoding. And thanks, will check out your thread on UHD 770 and subtitles. I figured worse case scenario (until bugs are sorted out), the i9 can brute-force software transcode (~60,000 Passmark) than my Xeon (~17,500 Passmark).
My SuperMicro (12-core Xeon D-1700) is a fanless small-form factor server; I’m sure an Nvidia Quadro won’t fit. It’s great for what I wanted: homelab, devops playground, Docker, Proxmox, etc…
For most part, everything that I overbuilt at home, I’ve more or less fully used down the road.
crosses fingers
I thought I overbuilt my home [wireless] network (full Ubiquiti setup) and NAS (12 bay). But I’m fully using both of it. I’m only have 5 bays left on my NAS and I’m using 18/20GB drives, haha.
I’ve read subtitles (and also found out the hard way on my Xeon) tax the Plex host. I figured I want ample extra i9 horsepower for subtitles.
My SuperMicro has 128GB of memory and has Proxmox already setup on it. I also prefer the SuperMicro as it has IPMI so I can remotely turn it on/off. My day job is DevOps/SRE so I got that handled, haha.
I figured I’ll throw 16GB/32GB of memory in this Plex server. If things change, I can toss/sell the cheap memory and max out the memory (and NVME) for another Proxmox host. Or use it as a random Windows host.
Well, I’d be interested to hear how much you can get going at the same time if you decide to go the i9 13 route. Don’t forget to consider extra memory for a ramdisk if you go windows or tmpfs on linux so you don’t waste nvme cycles with the transcode files.
The official advice is not to use a ramdisk like /dev/shm – but lots of us do and it seems to work fine. If transcoding mysteriously blows up it’s something to keep in mind.
Ramdisk vs. SSD performance is negligible. Some report scrubbing (rewind/forward) is faster with ramdisk.
I was more interested in ramdisk for performance; less interested about using ramdisk to wear on the SSD. I was going to use a prosumer/NAS SSD (500GB Seagate Ironwolf 525 NVME) to mitigate SSD wear.
Plex docs say…
“Directory to use when transcoding temporary files for streaming. This is useful if your primary drive has limited space. The directory used (whether default or not) needs sufficient free space, roughly equal to the size of the source file of the transcode plus 100MB”
My source files are massive 50GB+ 4K Remux’s. If friends and family use my Plex server later, I don’t think using a giant ramdisk is cost effective or scalable.
I think I’m going to opt for 16GB (or 32GB) of memory. And 500GB NAS-grade SSD. At least a 500GB SSD can accomodate ~7 concurrent source temp files.
Yep, that’s what it says, but in my tests I was able to transcode/tone map multiple simultaneous 4k HDR streams on a system with 8 GB RAM. It even worked on 4 GB. Each operation consumed 700-800 MB of ramdisk.
I do not know why, but the official documentation does not tell the whole story.
I did notice that… I don’t recall my temp directory using more than like 10-15GB.
I’m going to cron a “du -s /transcode >> /tmp/transcode_dir_size.txt”, do some tests, “cat /tmp/transcode_dir_size.txt | sort”, and see how big this directory actually gets.
Question: Will moving to QuickSync help with Downloads speed? When I tried to 4K Blu-ray Remux and transcode to a reasonable size for a tablet, it took over an hour. I think I tried searching for that topic and couldn’t find any relevant info.
I got my i9-13900 server built. Couple quick observations.
Without devoting too much time to workaround it, I couldn’t get the UHD 770 (i915) detected under Ubuntu Server 22.04 LTS; it strangely picked up an a780. I installed Ubuntu Server 22.10 and it picked up the UHD 770 (i915) without a hitch.
I have 64GB of memory on this ITX board. I thought it would be ample ramdisk. Transcoding 4K Remux to 4K “High” created a 17GB file in ramdisk. As much as I want to use the ramdisk for kicks (and probably negligible performance), it’s probably not very sustainable when friends and family start using it. I used “Download” to test this. On my Xeon, I was testing 4K Remux to 1080 and it was “only” creating 8GB on ramdisk (duh); I should have better tested this before buying 64GB of memory.
As far as I can tell, hardware transcoding is working awesome (yay).
My Xeon took 1.5 hours to transcode 4K Remux to 1080 for Download. My i9-13900 took 23m to transcode 4K Remux to 4K “High”.
So far, I’m super happy with it! Personally, I’m quite happy picking the fastest UHD 770 out right now.
