rtpengine GPU support

[Walter Schober]

Hi!

Googeling rtpengine GPU support leads to more or less one single result:

https://www.sipwise.com/archives/products/rtpengine

-> Features -> Addon Modules: "GPU Offloading is supported".

No docs, no specs, no more information. So I just put in a RTX4060Ti, installed the drivers.

Now libavcodec-extra59 has a resolved dependency to all libcuda1, libnvduvid1, libnvidia-encode1.

Is that all to do? I found a rtpengine-gpu module reference, but cannot find it again ATM.

Is this a commercial module? 

How to utilize OPUS-AMRWB transcoding using GPU ressources? CPU is using 7% single core usage per call unfortunately.

Greetings!

Walter

FTR: Debian 12 Nvidia Driver Install w/ i915 onboard + RTX is successful following https://docs.nvidia.com/cuda/cuda-installation-guide-linux/

[Richard Fuchs]

On 14/12/2024 11.40, Walter Schober wrote:
> Hi!
>
> Googeling rtpengine GPU support leads to more or less one single result:
> https://www.sipwise.com/archives/products/rtpengine
> -> Features -> Addon Modules: "GPU Offloading is supported".
>
> No docs, no specs, no more information. So I just put in a RTX4060Ti,
> installed the drivers.
> Now libavcodec-extra59 has a resolved dependency to all libcuda1,
> libnvduvid1, libnvidia-encode1.
>
> Is that all to do? I found a rtpengine-gpu module reference, but
> cannot find it again ATM.
> Is this a commercial module?
>
> How to utilize OPUS-AMRWB transcoding using GPU ressources? CPU is
> using 7% single core usage per call unfortunately.

This is indeed a commercial module as there are no (usable) open-source
implementations of GPU audio codecs. (FFmpeg links to CUDA because of
video codecs.)

Currently only G.711 and Opus are supported. AMR would have to be
implemented, if even possible due to licensing and patent restrictions.
(Alternatively a possible solution would be to keep AMR on the CPU and
only do Opus on the GPU. Not sure what portion of the load of an
AMR<>Opus transcoder is on AMR and what portion is Opus.)

Cheers