Hello,
we are currently switching from LiveStream ober Websockets to WebRTC. But when enabling WebRTC the kit process immediately has a very high GPU load even if no stream is connected and no stage is loaded.
We see this von Create 2022.2 in Windows and Linux. Is this a known bug?
Thanks
Carl
Hello,
can anyone confirm this? As sonns as we enable WebRTC streaming the gpu(s) go to 100% load, even if there is no stage loaded and no stream connected…
This is a screenshot from our KitAgent (an own app similar to Kit Farm, that manages kit processes for our solution). The values are fetched with NVME…
WebRTC streaming is very important to us (we are developing a product configurator that uses remote raytracing for visualization)…
@mati-nvidia can you have a look at this… this problem is a showstopper for us. because we want to have as many render sessions per box as possible (of course :-))
Hi @mati-nvidia, please have a look at this ticket… von create 2022.3 this problem is still alive.
…currently our last showstopper.
Thank you for your help
Carl
HI Carl,
Appolgies on this very late reply. I am looking into this for this for the first time. Let me contact the relavant parties and ask a question about it. Thank you for your patience.
Hi @Richard3D,
thank you very much for your support. Can you have a loot at this too please?
Livestream with multi GPU does only use one GPU for encoding? - Omniverse / Developer - NVIDIA Developer Forums
The high GPU-load even in empty idle sessions is a problem, but this one is a real showstopper, because it limits the number of concurrent streaming sessions to that what one GPU could handle - regardless how many GPU the system has.
Thanks
Carl
Hi @Richard3D,
one other issue around “GPU-Load”…
When starting a Kit-Process (for example Create) without loading any stage and without using streaming… this process generates 15% (A40 unter Linux) GPU-Load in idle state… even with Eco-Mode enabled (but ok, there is nothing to render because there is no stage loaded, so I think Eco-Mode makes no difference is this situation).
Here is a screenshot from our agent…
This process runs for some time now, but is completely idle (no stage loaded, no streaming)… but generates 15% GPU-Load. We would like to pool those render processes through our agent to react even faster, but when those idle processes produces such high GPU load, that pooling does not make sense.
PS: This problem occurs on Linus and Windows systems
There is some inherent baseline GPU usage to using the App, even if there is nothing loaded. It is the GPU essentially rendering nothing at 120fps. That is to be expected. It is not really a bug, as much as the way the real-time system works. You have to refresh your scene, empty or not.
I am assuming you have read through this manual and the links it provides…
https://docs.omniverse.nvidia.com/prod_extensions/prod_extensions/ext_livestream/webrtc.html#webrtc-browser-client
I think stopping the render loop when there is not a stage loaded shouldn´t be a big deal… but anyway. This is the smalest problem… the high load when enabling WebRTC is another thing… and that the system is only unsing the encode of GPU 0 is a real showstopper…
Yes of course I’ve read the streaming guid… but obviously it does not contain anything related to my problems…
Carl
Hello @c.bickmeier! I’ve sent this issue over to the development team to evaluate and fix!
A development ticket was created from this post: OM-77406: Very high GPU load when enabling WebRTC
1 Like