Previously, when running a certain ROS program, it utilized all 12 CPUs, which led to high system load. So, I decided to use the GPU to run the program instead. However, I’m wondering why the CPU usage hasn’t decreased
Which JetPack SW? CUDA version?
How does your program work?
How do you implement to use GPU to run program?
Any profile log as reference?
Your application may contain some parallel computing implementation.
So it’s trying to use the hardware as much as possible.
Please share more about your use case so we can suggestion further.
I used JetPack 5.1.2 and CUDA 11.4 version.
I utilized an already implemented ndt registration algorithm called ndt_omp.
The part of the code where parallel computation is carried out seems to be represented by the following code:
“#pragma omp parallel for num_threads(num_threads_) schedule(guided, 8)”
In the case of desktop ROS1, when running this algorithm on a GPU, the CPU is hardly used, but on Orin, running the same algorithm results in very high CPU usage
Could you help to confirm if the GPU task indeed runs on GPU?
For example, when running inference with PyTorch.
If the package doesn’t support CUDA, it will run the inference on the CPU instead.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.