# Why is a Nvidia Single Board computer less than ideal for AI model training?

Conventional NVIDIA GPUs, such as the Titan and or GT1080, are used to train AI models. Why would a Jetson Nano board be less than ideal as a substitute for a conventional GPU?

# CONTEXT

I would like to run simple concept demonstrator training exercises with and without a GPU: that being said, I am not interested in buying a desktop and outfitting it with a $1000+ GPU for said exercise. • Answered lol not sure why you edited this – hisairnessag3 Apr 5 '19 at 11:01 • Anything else you are looking for out of an answer??? – hisairnessag3 Apr 5 '19 at 12:14 ## 1 Answer Jetson Nano specs Price:$99:

128-core Maxwell GPU

Quad-core ARM A57 @ 1.43 GHz

472 GFLOPS max compute

Titan V(for example) Price:\$2500 :

640 tensor cores

5120 cuda cores

110000 GFLOPS max compute

Some simple maths gets us to a speed of 11,800 GFLOPS for 25 of these perfectly synced in parallel(not likely or gonna be fun). Vs the 110000 it just isn't even a question.