Off load task from CPU to GPU

Do we need to change the language.?

If you want off load the CPU load to GPU, first may need to profile the python function and may need implement by the same by CUDA

Can you suggest a good profiling method other than nsight systems, which won’t affect the working of main code.
We did the nsight profiling and it affects the working of main code. I don’t want this to happen in profiling.

Not sure if it’s the same with nsight.

We get to know that it works only for high end boards like TX1, TX2,.etc

Does it support on Jetson nano.?

Yes, it’s support Nano. Nano is the same with TX1 chip.

Supported target platforms:

  • Jetson TX2 Developer Kit
  • Jetson TX1 Developer Kit
  • Jetson TK1 Developer Kit

My PC is ubuntu 20.04, is it supported.?

Sorry, I didn’t run it before. Could you try it or have virtual machine to run it.

I’ve seen the YouTube sample of l4t profiler, this also works over the WiFi. So I’m predicting I’ll be experience the same issue as that of nsight systems profiler does. As it follows the same as that of nsight. So does it works in our scenario.?

Do you have any suggestions to the above.?

Sorry, I don’t have much experience about those tools.

Ok, can you say any sdk or provide any documentations for offloading tasks from cpu to gpu.

I think you need to figure which python element acquire much CPU resource then check if any GPU element to replace it.

Python element ?

I mean you need to know which python function consume much CPU resource and check if any alternative GPU solution to replace it.

How can we evaluate that.?

You need breakdown your python code.

We are using a camera streaming and displaying code. How should I breakdown our python code.?

OK, so you did use python?
But from your early comment python take 81.7%(20.5%) of CPU usage. And nvargu-daemon(camera) take 56.9%(10.6%)

Yes, we use python.
We are using cv2.imshow(camera_frame1) and cv2.imshow(camera_frame2)
In between imshow’s , we included a delay cv2.waitkey to make it display alternately.
So we are experience a lag in between these imshow’s.
Can you comment about our python function.?