Using CUDA unified memory decreases FPS rate

Accelerated Computing Intelligent Video Analytics DeepStream SDK
, ,
1

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU): GPU
• DeepStream Version: 7.0
• NVIDIA GPU Driver Version (valid for GPU only): 551.61
• Issue Type( questions, new requirements, bugs): bugs
• How to reproduce the issue ? (This is for bugs. Including which sample app is using, the configuration files content, the command line used and other details for reproducing)
• Requirement details( This is for new requirement. Including the module name-for which plugin or for which sample application, the function description)

Putting the following code to use CUDA memory in order to be able to access frames and save them within the pipeline lead to drastically decreased FPS and 100% GPU utilization:

        if not self.platform_info.is_integrated_gpu():
            mem_type = int(pyds.NVBUF_MEM_CUDA_PINNED)
            self.streammux.set_property("nvbuf-memory-type", mem_type)
            if self.platform_info.is_wsl():
                vc_mem_type = int(pyds.NVBUF_MEM_CUDA_UNIFIED)
                self.converter.set_property("nvbuf-memory-type", vc_mem_type)
            else:     
                self.converter.set_property("nvbuf-memory-type", vc_mem_type)

I’ve removed this piece of code and FPS went up to 25 from 2 (there were 6 cameras). And GPU utilization went to ~5%. Please, help me to resolve this problem1

3

which sample are you testing or referring to? are the sources the same RTSP URL? if using one source, will the issue persist?

4

It’s based on my own code, which is based on deep stream multistream app. When I launch it on one source it runs with 11 FPS, which is low. I can share piece of code if you need.

5

can you use deepstream-test3 to reproduce this issue? if using NVBUF_MEM_CUDA_UNIFIED, can fps reach 25?

6

I did this and fps can’t reach 25. It’s 16-17.

7

I modified the test3 with adding the code you shared. and I can’t reproduce this low fps issue. so mem_type is not the root cause. please refer to the code and log.
deepstream_test_3.py (18.4 KB)log-0829.txt (296.9 KB)
is there any mosaic on the video? please refer to this topic . is it related to the network packets loss?

8

I don’t think that’s due to internet quality. By the way, I’m running it on WSL and the problem might be in it. I can’t see reasons why would GPU utilization on one source go up to 100%.

9

what is your GPU device model? could you narrow down this issue? for example.

  1. if using local file sample_720p.mp4, will GPU go up to 100%? wondering if it is related to source.
  2. you can link nvstreammux to fakesink to check if the GPU can reach to 100%. then link nvinfer to fakesink if the GPU can reach to 100%. wondering which plugin caused 100% GPU.
10

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

11

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