Anyone have been able to HW transcode on an Intel NUC 11 (Iris Xe)?

I’ve been using a Samsung 980 Pro 1TB with no problems for the last few months on my PAHi7 (latest BIOS).

Had Proxmox running for a bit, now it’s running Ubuntu 21.04 with all updates, not noticed any problems at all (with the drive). Did a bit of Chia plotting as well - again, no problems noticed.

What kernel version are you running? I see 5.13.2 has dropped with a pretty hefty changelog.

@jmt089 @triks_melb Tried both. 4k hvec 10 bit transcoded, as did 1080p hvec and x264. However, it has proven to be very unstable. Sometimes it’ll chug along for an hour and a half, other times it’ll crash immediately. The only fix is a reboot, and the effectiveness of the fix is 50/50 (and temporary…). My system shows Ubuntu 5.8.0-59.66~20.04.1-generic 5.8.18. I guess the question is whether I can stomach starting all over again and trying the latest kernel.

Two of the test runs were a bust because of the drive problem. Thanks @DarthBJW for the tip! I’ll try flashing the bios again. When the transcoding is working, I find I’m usually able to skip around. Will look at the logs but I’m guessing it’s just going to show the same problem as everyone else here. I almost wish I could return the NUC. I’ve had less trouble with my old Celeron Synology that I was replacing!

1 Like

Well 5.13.2 was a disappointment. While there were some drm and 915 related fixes in the changelog, nothing to address our problem and it behaves exactly the same. Clearly we haven’t made enough noise :smiley:

@geeooff seems to have found the best outcome. I also tried enabling HW decoding and turning off encoding and it was a decent result. Transcoding a HEVC 1080 → 720 worked well and used about 20% CPU. Transcode of HEVC 4k → 1080 used 40% CPU, but it was buffering a lot. This may be down the to the fact I am mounting the media over NFS from a server with USB attached drives.

1 Like

If PMS is correctly utilising Intel quick sync or similar, 40% CPU on the i5-1135G7 which scores 10,030 would mean it most likely wouldn’t take much i.e. 4K 10bit HDR @ 30Mbps to max it out.

Doesn’t seem like the i7-1185G7 @ 3.00GHz would be worth it either as that scores only 10,958.

Plex recommend:

4K HDR** (50Mbps, 10-bit HEVC) file: 17000 PassMark score (being transcoded to 10Mbps 1080p)
4K SDR** (40Mbps, 8-bit HEVC) file: 12000 PassMark score (being transcoded to 10Mbps 1080p)

For some reason I thought the quick sync would leverage the CPU more.

This may be of interest to anyone still deciding on which PMS hardware to get:

Best PMS Server - Mac mini M1 vs NUC 11 i5/i7

@triks_melb I believe the 40% figure is with hardware encoding disabled–at least that’s the case for me. I was just now testing a 4k HDR hvec file to 1080p (20mbit “Univeral TV” setting) with hardware decoding and encoding both enabled and it was converting at between 3x-5.2x speeds and CPU usage held under or around 10%.
…I say “was” because one third through it crapped out and reverted to software. ( also don’t know if it attempted tone mapping even though the setting was enabled). But it indicates pretty good performance if this issue is ever sorted out.

I’m running 5.11.0-22-generic.

One of you guys needs to try Debian 11 - they build/maintain their own kernel based on 5.10 - just curious if it may work better than Ubuntu.

As stated a few posts up, the problem is not a flavor of distro but rather the Intel guys not getting their i915 driver fixed for TGL/RKL iGPUs. Didn’t find anything related/noticeable in the 5.13.3 changelogs.

Just in case any Intel and/or Plex folks (whoever can sort it out, I guess) are browsing this thread, another error I found is that hardware encoding+decoding+HDR tone mapping produces the following beauty:

Similar to the error documented here in March, but it only occurs when all three settings above are enabled.

I still think it is worth a shot. What’s the harm at this point?

I’m a recent victim to this issue after adding a few NUC11’s (i5-1135G7) to my cluster. I’m a bit disappointed now that I’ve read this thread in entirety, and the only functional solution is to run Arch. There’s also a bit of contention in my mind with this option. It’s been stated that it’s an upstream i915 driver issue. If that’s the case, why would Arch be functional? Has anyone else validated @Atlantis5’s approach?

just looking at these, with 23,298 PassMark could this be a better alternative to the NUC…

AMD Ryzen Mini PC

No, because Plex has no support for hardware encoding with AMD GPUs.

thats really unfortunate but thanks for the informative response.

I tried. God’s honest truth, I tried. But this is the first time I’ve been wading this deeply into linux and I was unable to successfully compile the driver they referenced on the Arch system. I did manage it in Ubuntu (the Arch driver referred to a git page with an Ubuntu build) but the result was no better, and in fact less stable for me. (Of course, that’s a different kernel version than Arch). I’ve since gone back to my previous Ubuntu setup.

Hmm never seen this issue .

Running with an Samsung Nvme 1TB Pro

@DarthBJW @turfrider @ChuckPa I’ve given a crack at making some more noise to hopefully get the attention of the Intel/i915 driver team, this time looped in a ticket manager and some Intel engineers I noted had been involved in resolving some closed i915 Gpu hang tickets within the past 12 months.

All, if you’re following this thread and are hoping to achieve a NUC11 or Tigerlake based Plex server at some point, please do us all a solid and share your issues here: https://gitlab.freedesktop.org/drm/intel/-/issues/3398

Alternatively you could also raise your own ticket, but I feel we would likely get the best response if there is substantial traffic on the one ticket. (See How to file i915 bugs · Wiki · drm / intel · GitLab)

Fingers crossed.

Oh that was you? Nice work!

I’ve added a comment offering testing support. Hopefully those you have tagged can become involved in this.

1 Like