Jetson TX1 having extremely low free memory despite having only one program running?

I am currently running a tensorflow program to test if it works, and I’ve gotten to the part of actually running the program. However, the program keeps on giving me an output called “Killed” and surprisingly have a very low free memory (1.7GB) although the total memory is around 4GB.

Furthermore, although the CUDA libraries have been loaded when I run the program, the fan isn’t spinning and I cannot use nvidia-smi to check what driver has been installed. Neither could I find the appropriate driver for the native Tegra X1 (if this is correct for TX1) if the driver is indeed absent.

What is happening in my program and is this normal? Do I have an nvidia driver installed if I could not do nvidia-smi? Because my TF program seems to be using the CUDA library.

Thank you for your help!!

You can use /home/ubuntu/tegrastat to monitor memory and GPU usage on Jetson TX1.

Reference: https://devtalk.nvidia.com/default/topic/928386/gpu-usage-monitoring-on-tx1/?offset=2

The most likely reason for your TensorFlow program getting killed is indeed due to the system running out of memory. Adding a swap partition might help in some cases.

Hi,

Thanks for your question.

Nvidia-smi doesn’t support Tegra device. Please use deviceQuery, one of the CUDA sample code, for system report.

If you want to enable fan, please run this script:

sudo ./jetson_clocks.sh

Tensorflow takes a lots of memory. Please add some swap space first to make it work properly.
Here is some information for your reference:
https://github.com/tensorflow/tensorflow/issues/851

Thanks.