I’ve replied with mine to the other thread. Tested outside of the docker environment to avoid any complications with this. Was there anything else you needed?
Now that we have 5.18, any updates on plex support?
I don’t know what Engineering might have done with the transcoder for the Plex Pass binaries.
What I just did for you and all is to make a PMS 1.27.0 package and place it in the new packaging thread for you to download if you wish.
At the bottom of my OP..
1 Like
i3-12100 | Ubuntu 22.04 | Kernel: 5.18.2
To jump in here, I gave the 1.27.0 package you linked to and still have the decoder failure when doing HDR tonemapping, obviously the SDR stuff is all fine.
I guess we’re just waiting on plex’s use of ICR to be updated?
If relevant, here’s some of the installer’s output:
RequiredIntelGmmlib="21.3.3"
RequiredIntelIGCCore="1.0.9441"
RequiredIntelIGCOpenCL="1.0.9441"
RequiredIntelOpenCL="21.49.21786"
VersionIntelGmmlib="21.3.3"
VersionIntelIGCCore="1.0.9441"
VersionIntelIGCOpenCL="1.0.9441"
VersionIntelOpenCL="21.49.21786"
Platform="B660M-ITX/ac"
Processor="Intel(R)"
Distro="Ubuntu 22.04 LTS"
Kernel="Linux raptor 5.18.2-051802-generic #202206060740 SMP PREEMPT_DYNAMIC Mon Jun 6 07:46:45 UTC 2022 x86_64 x86_64 x86_64 GNU/Linux"
Can you show screenshots of how it fails?
And, if you have DEBUG log files of that playback, they would be greatly appreciated.
Failure Screenshot:
Logs:
Plex Media Server Logs_2022-06-09_13-58-21.zip (473.2 KB)
debug_output.log (184.3 KB)
First time collecting the logs, so let me know if I didn’t grab what you needed. debug_output.log is simply the DEBUG messages from the console at the start and stop of this example. I did also try playing this back from a native player (e.g. iOS) and it still did this.
I will also corroborate that adding PGS subtitles removes the artifacting from the output, although on these bigger files it seems like NVENC and Quicksync (no Nvidia GPU in this system) both have a bottleneck somewhere that sits the CPU around 40% and doesn’t transcode fast enough to feed the player.
Thank you for this.
I will submit to the transcoder team.
When adding PGS subtitles, the hardware encoder is no longer used.
PGS subtitles require the CPU to burn the the subtitles into the image.
PMS does the encoding in software.
This confirms the hardware encoder isn’t operating as it should.
Personally, I have and use a Nvidia P2200 and avoid the Intel Quicksync problems.
When you use PGS subtitles with Plex, you need to have a CPU which is strong enough to burn those subtitle images into the video in real time speed
i3-12100 | Ubuntu 22.04 | Kernel: 5.18.2
Yeah, NVENC is certainly more fire and forget on the configuration side.
Not to be off topic, but the CPU is strong enough to transcode the files when PGS subtitles are not being burned, but CPU utilization drops when PGS subtitles are requested. Is that expected? It doesn’t seem to make much sense to me, but it seems like PGS stuff can be a headache anyways.
This is on the same system, but I’m using the latest plexinc/pms-docker image with the hardware encoder isolated away from it. Obviously the player here (Firefox) doesn’t support native PGS viewing.
I’m not really looking for a solution to this if it’s expected, just something I thought was odd.
I have not had opportunity to gauge performance of AlderLake at all.
I know the new 5.18+ kernels have the advanced support for core affinity management in AlderLake.
Normally, when you add PGS, utilization would increase however, (TOTAL GUESS on my part here), might the kernel itself have migrated the process to one of the strongest cores?
At the moment, I don’t know when I will have access to an AlderLake to profile it properly.
Well, I’m not sure how familiar you are with the P-Core, E-Core and Thread Director stuff, but these SKUs (up to the i5-12500) don’t have the Thread Director or the E-Cores, so they’re basically just “normal” CPUs.
I would be surprised if there would be kernel issues with this configuration, since there isn’t all of the extra stuff to manage.
As such, it sounds like something is getting screwed up here as well, since the idea would be for utilization to increase. If there’s something that’s useful for profiling, I can always run it on my end.
Intel Ark: Intel Core i312100 Processor 12M Cache up to 4.30 GHz Product Specifications
Thanks for sharing that info. Now you see how little I know of this new CPU.
I’ll likely have some access to a DragonCanyon NUC later this summer 
Core-related set aside. The AlderLake changed a lot internally from the ASIC perspective.
From what I’m told,
- PMS does work properly up to, and including, the -11xxx CPUs
- Engineering has pinned Intel Compute Runtime, required for HW ToneMapping., to version 21.49.21786
If you would like to conduct an experiment, I would appreciate hearing the results.
- Settings - Transcoder - Show Advanced – TURN OFF HDR Tone Mapping.
- SAVE the change
- Now test and observe the results
what I think you’re seeing is Software Tone Mapping AND Software PGS subtitle burning but without a machine this is just an academic guess
Yeah, I can imagine Alder Lake changed up a decent bit more than just the core
architecture/thread scheduling. I think it’s interesting that the 11th gen CPUs work properly, but the 12th gen ones don’t given that they have the same iGPU (at least the silicon, maybe not so much the interface).
I’ve gone ahead and run that experiment with the 1.27.0 build and disabled hardware decode/encode in the transcoder settings as well.
All four of these were run on the same portion of the media file for 2 minutes each.
Interestingly, No Tonemapping & No PGS manages to get past the 60 second buffer, so the CPU utilization dips near the end, and enabling Tonemapping & PGS individually still allow the CPU to keep up with playback, but do reduce overall CPU utilization.
Using Software Tonemapping & Software PGS starts buffering during playback (about .6x speed on the encode side), and has the least amount of CPU utilization.
I guess we’re just playing the waiting game still on support then?
If getting a machine is the only bottleneck, I’d be happy to sacrifice this one to science for a week or two as I have some spare SSDs to load a separate OS on. The 12700/12900 in those Dragon Canyon NUCs are gonna be a lot faster than the i3 though.
With DEBUG logging enabled, (I need this)
Please do one final run with HW acceleration enabled.
When complete, Download the logs ZIP and attach so I can see what the transcoder did on each run.
You want HW acceleration enabled with which Tonemapping/PGS config? Presumably just Tonemapping?
Outside of HDR content, are you having issues with HWA transcoding with SDR content using your Alder Lake setup?
1 Like
Nope, the HWA transcoding works like a charm when tonemapping is disabled or SDR content is used. Enabling the tonemapping gets the artifacting shown here: Plex Media Server on Ubuntu 21.10 with Intel 12th Gen (Alder Lake) - #46 by reptilicus
I don’t expect that to be resolved until the Plex developers update PMS so that it works with API v12 of ICR. If you take notice at Intel’s ICR Github, you won’t see Alder Lake listed as production until version 22.04.22286. I am not surprised that HWA tonemapping HDR10–>SDR via OpenCL malfunctions with ICR 21.49.
Yeah, that’s what I noticed from some other comments I’ve seen around here and seems pretty straightforward. Software tonemapping+pgs is a separate “issue”. I haven’t seen an answer for why CPU utilization would be going down when those are enabled in conjunction with one another.
I’ve seen a lot of talk about artifacting and bugs on hdr tonemapping. But is anyone else seeing total system crashes too? I have the i7* 12500. I tried hw encoding streams. They seem to run fine but either after a couple of minutes or after too many concurrent transcodes (last time it crashed at 8), the whole system pretty much crashes and I need to restart it.
I run ubuntu 22.04 kernel 5.15.0-37.
I am not sure what causes it right now, so I am interested if anyone else is seeing similar behaviour?
And if yes, is there a recommended setup (docker image, kernel and version) to fix this?
edit
i5 12500*
