Confusion about which one to choose jetson xavier nx 16gb ddr4 variant or jetson Orin nano 8gb ddr5 variant

[mrinmaydas2604]

I am planning to make and run a computer vision project on a jetson , so which one should I choose, actually I want to run multiple 1080p video anomaly detections on it(2 to 3) simultaneously , so I am confused which one of them will be best suited for my purpose jetson orin(8gb ddr5) or xavier(16gb ddr4).

[andres.artavia]

Hi @mrinmaydas2604,
From my experience ml models and overall machine learning tasks are memory intensive. Your requirements on memory depend on the type or preprocessing, model(inference) and postprocessing you are going to use. And also highly dependent on the actual processing pipeline implementation, ie using zero copy buffers or using copies to manage data between cpu/gpu and so on. Since you mention mutiple 1080p streams its recommended to go with the higher memory one(16GB). But if your model and processing pipeline are lightweight you can go 8GB.
So at the end of the day, the benefit you can get by the ddr4 to ddr5 speed increase can be overshadowed by the data not fitting in memory. So maybe run a quick PoC on any other Jetson or PC to try and get an approximate memory usage and that way you can get a better idea of what you need.

Regards,
Andres
Embedded SW Engineer at RidgeRun
Contact us: support@ridgerun.com
Developers wiki: https://developer.ridgerun.com
Website: www.ridgerun.com