Help please! - Image prcoessing in Jetson

Autonomous Machines Jetson & Embedded Systems Jetson Nano
1

i don’t know how to solve this problem, please help me

  1. We use Jetson GPUs for image processing.

  2. But that processing, frames are so high and the image size is also large

  3. So then, It takes some time to move the memory of the CPU where the image datas are stored to the GPU.

  4. As I said before, I’m worried about that a bottleneck problem will occur in the process.

is there a way or another way to solve the problem?

ex) Way to receive data directly to the GPU without going through the CPU.

ex) Now we use cudaMemcpy method to transfer data from CPU to GPU.
Is there any faster method than this?

3

Hi,

You can try to use unified memory so the copy is not needed.

Thanks.

4

Hello AastaLLL,

could you develop your answer ? I am also interested in memory management with Jetson.

With my software (JetsonSky using Python and Pycuda), i have many data transfers between CPU and GPU.

Alain

5

Hi,

You can find a sample below:

/usr/local/cuda/samples/1_Utilities/UnifiedMemoryPerf

Thanks.

6

Thanks !

I will look at it.

Have a nice day.

Alain

8

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