Unable to use hardware acceleration (nvidia gpu)

[gnoacco]

Hello everyone,

I use NoMachine with Windows based servers since a few years now (using it to remote edit videos, etc, works great). The machine I’m using as a server is equipped with a Nvidia GTX 1050Ti GPU, working well with NoMachine (providing hardware encoding of the session).
However, I’m investingating an evolution of this server to linux (Debian 10) for my workflow.

I’ve configured Debian, installed a few packages (nvidia-driver nvidia-settings nvidia-xconfig libnvidia-encode1 libnvcuvid1), everything works well with obs-studio, blender, and a few other software. Running simple tests like glxgears and nvidia-smi shows the GPU is correctly accessed/used by programs, so everything looks fine here. Even OBS-Studio is able to access the Nvenc hardware encoder, to save/stream video on this machine.

I’ve installed NoMachine 6.10.12 on it, I can connect from other computers (running NoMachine Enterprise Client 6.10.12), but hardware accelerated encoding of h264 is not working: high CPU usage in htop, nvidia-smi shows the GPU is used by Xorg (or other programs, but not nomachine, nxnode, etc) and the NoMachine session logs on the server shows “Info: Using H.264 software encoder.”

Client: Debian 10 (amd64, LXDE, fully updated), NoMachine Enterprise Client 6.10.12 .
Server: Debian 10 (amd64, LXDE, fully updated), GeForce GTX 1050Ti (non-free nvidia-driver 418 – also tried with 440) , NoMachine 6.10.12 .

Have you already encountered this issue ? Am I maybe missing some packages/dependencies ? Is there an incompatibility between NoMachine and this config ?
I’ve read the documentation, and understood hardware acceleration was available for Windows and Linux servers.
I can use it with MJPEG, but colors are off and it generates far more network traffic as h264. So any help would be appreciated 🙂

Thanks a lot

[fra81]

Hi,

can you send the whole log where you see “Using H.264 software encoder”? You can pack the whole log directory as explained in https://www.nomachine.com/AR10K00697 and send to forum[at]nomachine[dot]com, if you prefer.

[gnoacco]

Hello,

Thanks a lot for your answer.
Here are the logs

Attachments:

1. NoMachine-log.tar.gz

[fra81]

Hello,

please run this command and show the output:

sudo find / -name libnvidia-encode.so -or -name libcuda.so 2>/dev/null

[gnoacco]

Hello,

Here is the output. It found 3 files:

/usr/lib/x86_64-linux-gnu/nvidia/current/libcuda.so

/usr/lib/x86_64-linux-gnu/nvidia/current/libnvidia-encode.so

/usr/lib/x86_64-linux-gnu/libcuda.so

 

[fra81]

It’s not clear why it doesn’t work. Would you be available to install a debug library from us?

[gnoacco]

Yes, sure !

[gnoacco]

So, where can I find this debug library ?
Thanks

[Britgirl]

Hi, can you check your email inbox because I sent a link yesterday with instructions. I’ll send it again.

[fra81]

It looks like libnvidia-encode.so can’t be loaded. The possible cause is that /usr/lib/x86_64-linux-gnu/nvidia/current/ is not in the library search path. You could try to create a symlink to that library:

sudo ln -s /usr/lib/x86_64-linux-gnu/nvidia/current/libnvidia-encode.so /usr/lib/x86_64-linux-gnu/libnvidia-encode.so

If that won’t work, please try to check permissions for that library and other libraries linked to this one (you can find such linked libraries with the ‘ldd libnvidia-encode.so’ command).

[gnoacco]

Your solution works !
The GPU is now acclerating the NoMachine session.

Thanks a lot !

[Author]

Posts