#Is there a way to see if nvenc is used for transcoding?

1 messages · Page 1 of 1 (latest)

quick mason
#

I downloaded all drivers for nvidia on debian and I run immich in docker. I also selected Nvenc for transcoding, but I'm not sure if it is working or just fallback on the cpu. Is there a way to see if Nvenc is really working?

mild widget
#

You can check gpu utilization with nvtop. The server logs will also show errors saying it’s falling back to cpu if it is

quick mason
#

But to do that I need to trigger the transcode job, so record something random and push it I think. I'll do that and give you my logs

#

does that help you?

#

or is that machine learning and has nothing to do with nvenc?

mild widget
#

I don’t see a transcoding log there

#

ML is different

quick mason
#

will it shows there or do I need to go into the docker container in the logs?

mild widget
#

It would show there if it started transcoding. It could be that the transcode policy didn’t think it was necessary to transcode it. You can change that in the transcoding settings

quick mason
#

ok thanks, I'll try to force it, record and see

#

in the meantime, I have this error everytime I start with docker compose up:

#

do you know something about that?

#

@mild widget There's my output:

mild widget
#

Transcoding looks like it’s working fine

quick mason
#

(I've activated cuda for machine learning for your information)
I saw my gpu utilization go up to 5% so that would mean it's working

mild widget
#

Also make sure hardware decoding is enabled too

#

Something is going wrong for ml though

quick mason
#

how did you see that transcoding worked successfully? I can see the job started but it never said finish with success?

mild widget
#

Toward the bottom of that image it says it encoded successfully

quick mason
#

Decoded already activated:

quick mason
#

And thanks a lot for you help btw, I really appreciate

mild widget
#

It says it got code 139 at the bottom, so a segfault. I’ve seen a few people with that issue and think it started with onnxruntime 1.17.1. Not sure if there are any workarounds

quick mason
#

oh no 😦 So it has nothing to do with a bad installation of cuda driver or anything like that you mean?

mild widget
#

It might still depend on the driver environment. You can try updating the nvidia driver and nvidia container toolkit

quick mason
#

I just installed it this morning so I'm at the latest. Do you have a link with someone with the same bug?

#

I would like to read on that

#

Would you want me to create a help desk support ticket?

mild widget
#

Some discussion about it here #imagegenius-aio message, here #1251160799834144861 message, and in GH at #10343, #10376 and #10727

scenic joltBOT
quick mason
#

Thanks a lot for you help. I really love where the community for Immich is at. Bright future!