Is there a limit on the number of streams in deepstream 8.0?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

Please provide complete information as applicable to your setup.

• Hardware Platform： GPU（RTX5090）
• DeepStream Version： 8.0
• TensorRT Version： 10.9.0.34
• NVIDIA GPU Driver Version： 575.64
• Issue Type： questions

• CUDA version： 12.8
• When running more than 27 streams on a pipeline, the frame rate printed by nvdsosd is less than the original video’s 25 fps, only around 16 fps, at which point the decoder utilization is only 50%. However, when I reduced the number of streams to 18, the frame rate was able to reach 25 fps.

3

What is the GPU utilization when you run your app with 27 streams?

4 
About 40%
5

Can you tell us your complete pipeline and configurations?

6

This pipeline diagram，Because the test removes the subsequent components of the nvstreamdemux：

pipeline.zip (468.7 KB)

Configuration file：

ds_app_config.zip (645 Bytes)

config_infer_primary_yolo11_all.txt (927 Bytes)

7

Can you set the “sync-inputs” property of nvstreammux to “FALSE”?

8

What are the resolution and codec format of your rtsp streams?

9

We have tested turning off sync, but the situation is still the same

10

3840x2160，h265

11

If your rtsp streams are all the same, the 27 4k@25fps HEVC streams decoding may exceed the hardware limitation of RTX 5090.

12

However, the decoder usage is only about 50%. Does this limit refer to quantity? Is there any official explanation from 5090?

13

How did you measure the decoder usage?

14

Use the terminal command nvidia-smi dmon

15

Currently no specific codec performance data for RTX 5090. It has two 6th generation decoder cores. GeForce RTX 5090 Graphics Cards | NVIDIA

The 4th & 5th generation decoder core performance data is available in Video Codec SDK | NVIDIA Developer.

You can also test the decoding only case to verify the decoder performance.

16

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.