Very recently I've started to notice that any 4k HDR source that transcodes (and tonemaps because, yes it needs to) will randomly crash the encoder and fallback to CPU.

This came up trying to troubleshoot another issue in that I have an AMD iGPU that was caught in the dri mapping snafu. You thought you were using the Intel when in fact you were running the AMD. Result=glitches, corruption just unwatchable.

I've been going down a rabbit hole trying to get HDR tonemapping while transcoding workg again and every approach I go down has been a failure. I now come to the understanding the using the XE driver can't do tonemapping at all. It can transcode down in scale but can't tonemap. If that's on, it straight to the CPU without even trying.

What does Unraid need to have in order to get this working?!

And I am floored that I don't see anyone else talking about this, unless I'm the only one that has something borked!

Have you considered filing a bug report with the transcoder software developer? (plex, jellyfin, emby, ffmpeg)

I spent the last 4 days doing 4K HEVC HDR -> 1080p SDR tonemapped transcodes on my intel A310 GPU no problems using plex.

Edited Wednesday at 10:07 PM1 day by MowMdown

I figured I'd ask here first in case anyone else has run into this problem and found a solution.

Let's compare notes; I'm running a smaller AMD 8500G on an Asus B650M board. A380 is in slot two, limited to 1 PCIe lane, unfortunately. The disk controller is in slot one for the lanes.

I mean Im running a 5800X (no iGPU) with a A310 in a gen3 x 1 lane PCIe slot as well because video encoding doesnt need much bandwidth.

There are a bunch of folks in the unraid discord server that also run intel iGPUs and I have not heard anyone complaining about this but I will keep an eye out.

Edited Wednesday at 10:41 PM1 day by MowMdown

Side quest @MowMdown , what idle power consumption fo you see on that a310? I have the full size single slot Sparkle version and best idle I can seem to get is ~11w.

***UPDATE: Downgrading Plex to tag 'plex:1.43.1.10576-06378bdcd-ls300' has resolved the tonemapping via OpenCL so all is back to working again on i915 with the A380. Tested with 5 4K HDR tonemapped 1080p streams. Flawless with headroom.

They've broken the pipeline somewhere since then

Careful you don't get bit by the built-in update catch! If you have an ENV var set to "latest", Plex will update itself inside the container, so it won't matter what image you point at. Set the env var to "docker"; that will lock the Plex version in the container to the version specified by the image tag.

I'm running the Linuxserver image, so the full image is lscr.io/linuxserver/plex:1.43.1.10576-06378bdcd-ls300

Edited yesterday at 12:11 AM1 day by aglyons

55 minutes ago, Veah said:

Side quest @MowMdown , what idle power consumption fo you see on that a310? I have the full size single slot Sparkle version and best idle I can seem to get is ~11w.

GPU stats doesn't report wattage so I have no idea

28 minutes ago, aglyons said:

***UPDATE: Downgrading Plex to tag 'plex:1.43.1.10576-06378bdcd-ls300' has resolved the tonemapping via OpenCL so all is back to working again on i915 with the A380. Tested with 5 4K HDR tonemapped 1080p streams. Flawless with headroom.

They've broken the pipeline somewhere since then

Careful you don't get bit by the built-in update catch! If you have an ENV var set to "latest", Plex will update itself inside the container, so it won't matter what image you point at. Set the env var to "docker"; that will lock the Plex version in the container to the version specified by the image tag.

I'm running the Linuxserver image, so the full image is lscr.io/linuxserver/plex:1.43.1.10576-06378bdcd-ls300

What version were you on before? Im on 1.43.3.10896 and it's fine for me.

16 minutes ago, MowMdown said:

What version were you on before? Im on 1.43.3.10896 and it's fine for me.

I was on the ":latest" tag forever, along with the "latest" env var, so it was always bleeding-edge, not a beta version just the very latest releases. I tried a few different tags before I hit this one, from I believe 8 months ago, that started working properly again. During this time, there were no definitive logs to show what was happening. All that would happen is the GPU would just stop handling the transcode, and it would silently flip to the CPU. I wouldn't even notice anything different about the stream. But the CPU would jump to 100% for a single stream. I should add that it also ONLY ever happened when tonemapping was involved. A straight 4k SDR to 1080p wouldn't crash at all.

Edited yesterday at 12:59 AM1 day by aglyons