Nvidia HW transcoding not working

[jcbdrs 0]

Hi everyone!

I am using Emby since years now and I am overall very satisfied with it.

I recently wanted to step up the game so I purchased a GTX 1650 to do some HW transcoding (I was using software transcoding).

So I did that, I followed all the guide I could find on how to install the Nvidia-driver than afterward the docket integration. Because yes my server was on Docker.

I could not get it to work, i had a lot of errors in the transcode logs, and under Emby --> Transcode --> Advanced - I had no options there to check.

So I found out that HW transcoding on docker was not really supported so I installed the standalone app with dpkg (the last version). What was my suprise when I finished the configuration (I started fresh on a empty folder) that I still had no options for transcoding with my Nvidia card.

When I check the hardware detection log, the errors showing there are exactly the same as the one I was getting with the docket instance.

The graphic card is working correctly, it displays the terminal when I plug it to my screen (since it's a server, there is no GUI and I am using the onboard VGA by default).

So whenever I run nvidia-smi I can see the version of the driver installed, my card is showing there as well and everything seems fine.

Here is a summary of what I tried so far :

-restarted the server numerous time.

-re-installed the drivers numerous times and from multiple sources, so either from the non free depot, from the Nvidia website or from their depot.

-re-installed fresh Emby.

-used the test docker on the nvidia-smi GitHub, and it was successful, so maybe it can be because the driver and CUDA version installed on my computer are too old for Emby?

-verified that the user who is running Emby is part of the video and render group.

-tried the beta version of Emby via both the docker and the standalone version, exact same behavior.

Thank you in advance for your help everyone!

hardware_detection-63728027283.txt nvidia-smi.txt

Edited June 18, 2020 by jcbdrs

[Luke 38803]

Hi, regarding Docker, did you take a look at this?

 

[Luke 38803]

Also can you please provide the emby server and hardware detection log from the native package?

@softworkz will take a look.

Thanks !

[softworkz 4278]

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.113      Driver Version: 418.113      CUDA Version: N/A      |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1650    On   | 00000000:82:00.0 Off |                  N/A |
| 35%   32C    P8    N/A /  75W |      0MiB /  3881MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

@jcbdrs - The installed driver does not include CUDA. Please install the latest drivers from the Nvidia website. 

[jcbdrs 0]

5 hours ago, softworkz said:

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 418.113      Driver Version: 418.113      CUDA Version: N/A      |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1650    On   | 00000000:82:00.0 Off |                  N/A |
| 35%   32C    P8    N/A /  75W |      0MiB /  3881MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

@jcbdrs - The installed driver does not include CUDA. Please install the latest drivers from the Nvidia website. 

Thank you for the info! 

I swear I have it installed at some point, but I tried so much things.

So I did not installed the version from Nvidia, because it is only compatible with Ubuntu and when I tried at the time it was giving me an error - probably because Debian is using older packages.

So instead I installed nvidia-cuda-toolkit from the non-free Debian repository.

Now when I do nvidia-smi, I can see that CUDA 10.1 is installed. I will restart and spin up a fresh install of Emby once everybody will be logged out from Emby.

I will also attach the server log if it does not work.

nvidia-smi.txt

[softworkz 4278]

It is not CUDA alone, so I'm not sure whether it will work.

PS: No need to re-install Emby

[jcbdrs 0]

11 minutes ago, softworkz said:

It is not CUDA alone, so I'm not sure whether it will work.

PS: No need to re-install Emby

Are you referring to https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&target_distro=Ubuntu?

Seems like it's similar to the package I installed, but a more recent version of it.

I gave up 2 days ago so I uninstalled the standalone app to use it in a docker instead so that's why I need to re-install.

Note : nvidia-driver is currently not installed, the toolkit seems to have installed some sort of driver, but not that one.

Edited June 23, 2020 by jcbdrs

[softworkz 4278]

Just install the latest graphics driver from Nvidia. Nothing else. No CUDA SDK or toolkit.

Edited June 23, 2020 by softworkz

[jcbdrs 0]

22 minutes ago, softworkz said:

It is not CUDA alone, so I'm not sure whether it will work.

PS: No need to re-install Emby

I think Emby is trolling me. It's all working now with the standalone app. Only thing left is to make it works on Docker!

Thank you really much! 

[jcbdrs 0]

1 hour ago, softworkz said:

Just install the latest graphics driver from Nvidia. Nothing else. No CUDA SDK or toolkit.

Thank you! 

I was actually testing the HW with the standalone app.

First, there is no hardware encoders options in Emby. I can only see the GPU under hardware decoders, is it a normal behavior? 

Second, whenever I start a movie (I tried with multiple format), it says in the dashboard that it is transcoding to software h264.

When I check nvidia-smi, I do see an Emby process there that is taking memory and GPU ressources. However, my CPU is also working as hard as per top.

I made sure to use a supported movie format (1080p h264 mkv) and I disabled the subtitle on my client.

[softworkz 4278]

1 hour ago, softworkz said:

Just install the latest graphics driver from Nvidia. Nothing else. No CUDA SDK or toolkit.

 

[jcbdrs 0]

7 hours ago, softworkz said:

 

I kept the config I had and I installed nvidia-docker. It's working inside the docker now.

Thanks!

[Luke 38803]

What about the native package?

[jcbdrs 0]

3 minutes ago, Luke said:

What about the native package?

The only thing I did different this time is I did not insta Nvidia-driver along side the toolkit and it worked. Seems like the driver is included with nvidia-smi and it was making conflict with nvidia-driver.

[Luke 38803]

Thanks for the feedback.

[softworkz 4278]

NOTE - For other readers

What is described above is NOT the correct way. 

Please simply install the official Nvidia drivers from the Nvidia website.

This has worked fine for every other user so far!

[Luke 38803]

@cayars will ensure our hardware acceleration guide reflects this guidance. Thanks.

[Carlo 4478]

Based on what softworkz said, nothing should need adjusting as he covered this in detail for those following the knowledge base.

[softworkz 4278]

19 minutes ago, Luke said:

@cayars will ensure our hardware acceleration guide reflects this guidance. Thanks.

 

7 minutes ago, cayars said:

Based on what softworkz said, nothing should need adjusting as he covered this in detail for those following the knowledge base.

@cayars - that is misleading. The right answer is: we have already worked together and updated the corresponding article in this context.

Edited July 6, 2020 by softworkz