Low Performance (FPS) between Deepstream + Triton architecture with Deepstream architecture only

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform (Jetson / GPU) : Jetson AGX Xavier
• DeepStream Version: 5.1
• JetPack Version (valid for Jetson only): 4.6

• Issue Type( questions, new requirements, bugs): Low Performance (FPS) between Deepstream + Triton architecture with Deepstream architecture only

• How to reproduce the issue ? ()
I make the performance TEST comparison between Deepstream + Triton architecture with Deepstream architecture only on Jetson AGX Xavier with the same video following these sample apps:

  1. Deepstream + Triton: Following the instruction here, I run the sample app with this command:
    deepstream-app -c source1_primary_detector.txt text

  2. Deepstream only: I run the deepstream-test2 sample from here with command:

python3 deepstream_test_2.py /opt/nvidia/deepstream/deepstream-5.1/samples/streams/sample_720p.h264

The result shows that Deepstream only is faster than Deepstream + Triton. So my questions are:

  1. Is Deepstream only is faster than Deepstream + Triton ?
  2. If NOT, do I miss something in installations ?

You can see the video that I record. First is Deepstream + Triton from [00:00-01:19] , Second is the left.

Thank you very much!

1 Like
3

Hi guys, Is there any update from NVIDIA ?

4

+1

5

i see in this document has say triton less performant than a TensorRT solution at Table 1. TensorRT and Triton Server trade-offs.

6

Thanks, I agree with you. But I need the confirmation from NVIDIA Engineer. Because it will affect our product architecture design.

1 Like
7

Sorry for delay!
Yes, if your model can run with TensorRT, we recommend to use DeepStream with nvinfer (TensorRT), it will be faster as per our experience.

8

Thanks alot

2 Likes
10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.