Deepstream NvDCF tracker on running on 2 GPUs

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform (Jetson / GPU) T4 GPU
• DeepStream Version Deepstream-5.0
• NVIDIA GPU Driver Version (valid for GPU only) 460.32.03

I have multiple GPUs on my instance. I am trying to run the whole pipeline on the GPU-id 1. Everything is working fine, but when I am using the NvDCF tracker I am getting a redundant process on GPU 0 (default GPU according to SDK) that is consuming 0% utilization of GPU 0. However, it is reducing the processing time of the whole video by a significant amount (for 15 videos - each is 10 min - and 3 models, running the whole pipeline on GPU 0 requires 6 min of processing while running it on GPU 1 with this redundant process consuming 0% utilization on GPU 0 the required processing time is around 10 min).

Moreover, this issue is appearing during runtime when all the components are created and the pipelines starts running.

Using KLT or no tracker, this redundant process is absent. This is affecting our scalability a lot. Can you plz follow up with the issue?

For you to replicate the issue, I modified deepstream-test1 to add a tracker, run on GPU 1 and output an mp4 file. Please find the link for code in the repository and attached is a sample output of the problem.

image
image701×545 13.3 KB

3

Thanks for the sharing, I will try locally to see if can repro it.

4

Please note that we did the same test using Deepstream 5.1 and the issue is solved.

Although the redundant process still exists, it is not affecting the processing time and load.