Jetson 2gb with Hello AI Course

I have been doing the hello ai world course on the nano 2gb and the RAM usage has been going up 100 MB less than the max RAM of the nano. I was wondering if with other models later on it could be an issue.

Thanks,
Joshua

Hi @josh123, most of the models from Hello AI World that I tested on Nano 2GB worked ok, except for VGG-16/19 I think (those are rather outdated models anyways). These should all work:

Alexnet
Googlenet
ResNet-18
ResNet-50
ResNet-101
FCN-ResNet18
SSD-Mobilenet-v1
SSD-Mobilenet-v2
SSD-Inception-v2
DetectNet

When the RAM usage approaches it’s maximum, that’s when the system starts to use SWAP or ZRAM (compressed SWAP stored in RAM).

Would the models work better on the 4gb nano? Also, if I wanted to combine ROS with some models would I run into issues?

Thanks,
Joshua

Once they are loaded, the runtime performance is the same as Nano 4GB.

I did some basic testing of ROS Melodic on Nano 2GB without issue (see the ros_deep_learning nodes for Hello AI World). Although if you want to use more memory-hungry ROS stuff like SLAM, RTABMAP, ect you may benefit from the extra memory.

1 Like