Processing on GPU

Hello, I am quite new to GPU processing and the Jetson Products.

I am currently training a machine learning model on a Jetson AGX Xavier, but it is taking way too long been waiting for about 24 hours right now. I just realized I might not be making use of the GPU, meaning that I am probably running on CPU that is why it’s taking so long.

My question then is how do I execute scripts on Xavier on GPU rather than CPU?

Hi,

In general, we don’t recommend Jetson platform for training.
Jetson is designed as an edge platform so the memory and bandwidth might not be the optimal for training task.

About how to use GPU, it depends on the frameworks you used.
For example, you will need to install tensorflow-gpu rather than tensorflow to get the GPU support.
https://docs.nvidia.com/deeplearning/frameworks/install-tf-jetson-platform/index.html

Thanks.

I am using pytorch, Is there a corresponding tool/package for pytorch?

Hi tangenishikomba, please see this post for the PyTorch packages for Jetson:
https://devtalk.nvidia.com/default/topic/1049071/jetson-nano/pytorch-for-jetson-nano-version-1-3-0-now-available/

Thank you!

Can you please advice me on which platform from NVIDIA would be suitable for training?

Thanks!

Hi tangenishikomba,

A host machine with NVIDIA dGPU will be more suitable than using Jetson platform on training.
You can also refer to below web page: https://www.nvidia.com/en-us/deep-learning-ai/developer/