Cannot run Always-On Sink adapter in Docker on A100 about savant

[bwsw]

Yes, you must disable gpu resources for this particular container.

[bwsw]

Or you can build your own AO-RTSP based on our container code.

[bwsw]

@patrickhulce hello WSL is not fully supported by DeepStream and, thus, Savant.

[bwsw]

Hello, read this, not a bug:

https://docs.savant-ai.io/develop/savant_101/10_adapters.html#always-on-rtsp-sink-adapter

[cookiego]

I suppose it seems not able to detect that A100 in place does not support hardware encoding.
I pull ghcr.io/insight-platform/savant-adapters-deepstream:latest images and i read in docs there was like a failback to software encoding, but in logs I see that it successfully detected hardware

[bwsw]

Here is the error. A100 does not have NVENC on board:

ERROR insight::savant::deepstream::encoding > You have configured NVENC-accelerated encoding, but your device doesn't support NVENC.

[bwsw]

https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new

[cookiego]

Hmmm, as far as I can see you are saying to me that I should disable Nvidia Runtime for that container. Ok I'll try that way, but I have a docker swarm configuration, so the runtime now is activated as a whole for the swarm, not sure about concrete containers.
I'll check and return.
Thank you!

[cookiego]

Here is the error. A100 does not have NVENC on board:

ERROR insight::savant::deepstream::encoding > You have configured NVENC-accelerated encoding, but your device doesn't support NVENC.

Sure i understand this. It is written in Savant's docs that A100 does not support NVENC. My question is that when i run docker container on A100 if fails to start (even it is supposed to switch to software encoding). But, yes, maybe my docker swarm config forces to use nvidia runtime\hardware encoding even if real harware does not support it

[cookiego]

Or you can build your own AO-RTSP based on our container code.

Yes, indeed, I'll try that.
Just for somebody with the same prob at this time. What i've found is that currently when you are in swarm mode of docker, there's not support for setting 'runtime' on per-container basis in your .yaml. So the only way to use NVIDIA\CUDA hardware support inside container is to set the 'default-runtime' in docker daemon.json, which in my case (starting AO-RTSP fails saying it has hardware support but it is busy, when there's no hardware support at all...)
So, I'll try to build custom thing.

Thank you!

[bwsw]

Hm, the problem is not about the runtime, but only resource deployment. Does not swarm allows that?

[patrickhulce]

I'm getting this same error on a Windows device in WSL even though my graphics card (GTX 4070) definitely supports NVENC (it does within WSL, but apparently these docker images can't detect / use it). Any guidance / troubleshooting advice here to debug further?

[patrickhulce]

Thanks @bwsw !