deepstream 6.2 dgpu rtx3070
Multiple sources multiple rtspsink 。
I pull 16 rtsp streams, and each rtsp stream is connected to an rtspsink. When I reached the 13th rtspsink, the video playback graphics froze. May I ask what the reason is and how can I troubleshoot performance bottlenecks
Is there any abnormal in the log when the issue happens? You can add “GST_DEBUG=3” before your command to collect more logs.
Not sure if some hardware component becomes bottleneck, you may use “nvidia-smi dmon” to check the usage of different components.
this is the nvidia-smi dmon displays:
I want to know if the bottleneck has been reached and where it is？
According to the nvidia-smi log you post, the GPU is overloadded.
I want to pull 50 rtsp video streams and play them through 50 rtsp sinks. Which graphics card can meet my needs? RTX3070 is determined to be unable to meet the demand, right？
What is the pipeline?
I only have one pipeline. All rtspsources are added to this pipeline and distributed to 50 rtspsinks using demuxer
Is the pipeline just as simple as
rtsp sources->video decoders->nvstreammux->nvinfer->nvstreamdemux->video encoders->rtpsinks->rtspservers ？
According to your nvidia-smi log, the hardware video decoder and hardware video encoder are very low usage. Have you monitored the CPU loading with the 16 rtsp inputs and 13 rtsp outputs?
The log shows the decoder is 100% usage.
Please refer to the hardware-accelerated video decoding capability in Hardware-Accelerated Video Decoding
Take the 1080p30fps as the example, L40 can support 67 streams decoding. So you need to choose the same capability GPUs for your case.
It is decoder usage. mclk and pclk are the last two items.
many thanks
I would like to choose a graphics card with high cost-effectiveness. Can A2 card decode 50 1080p videos simultaneously? Where can I check the decoding concurrency performance of different cards？
A2 is smilar to A10. Video Encode and Decode GPU Support Matrix | NVIDIA Developer
What is the relationship between Total of NVDEC and the number of decoded video streams?
With the same GPU family, the more NVDEC numbers, the more decoding capabilities.
A2 is the same “Ampere” GPU with 2 NVDECs as A10.
Thanks
I want to reduce the decoding pressure by reducing the frame rate of the rtsp stream. May I ask if it is possible and how can I do it？ thanks