After sometime FPS is fluctuating hit's 40 and came to 5 ! Asking for a solution

[debjit.adak]

Please provide complete information as applicable to your setup.

**• Hardware Platform ---------> GPU
**• DeepStream Version --------> 7.0
• TensorRT Version --------> 8.6
**• NVIDIA GPU Driver Version -------> 545

I am using 4090 as a VM, and running 40 cameras most of them are 25 FPS, some of them are 13 FPS and 20 FPS.
First few hours It’s stable after that it’s started fluctuating, time by time fluctuating is increasing a lot. I am pasting a FPS.csv file and marge FPS_with_nvidia-smi_dmon with timestamp. What I observe is dec is touch 0 and goes back to 100/above 90 , I am sharing all information and asking to solve this ASAP !!!
sharing_with_nvidia_forums.zip (818.7 KB)

Please suggest us some solution !!!

[fanzh]

which sample are you testing? how did you get the fps? are there RTSP reconnection? if yes, the fps will go down then go up.

[debjit.adak]

Hi @fanzh

I was doing our pipeline, pipeline element looks like nvurisrcbin —> streammux —>queue ----> nvinfer —> tracker ----> queue ----> appsink.

I am having nvidia-deepstream example FPS.py how does it doing we are using the same thing. No there is no re-connection happening of the camera.
Above zip file I share If you go through you can understand it. I am attaching the graph for your better understanding.

[fanzh]

1. how did you " there is no re-connection happening of the camera"? Thanks! please refer to this topic , if setting rtsp-reconnect-interval, the nvurisrcbin will try to reconenct the rtsp source. could you share a whole log deepstream? Thanks!
2. which plugin 's probe did you call “perf_data.update_fps” ? To narrow down this issue, you can simplify the pipeline. for example, remove temporarily trakcer, nvinfer step by step.

[yingliu]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks