the good and bad of k210 in my view
yan211 last edited by
the good aspect: cheap enough , the usage is easy and transparent, one can put data and param to get the result layer by layer, the calculation process and result can be exactly the same as they on pc deep learning tools by little modify code.
ram is scarce，only toy network can be feed. the main stream model is too weight to run. when use big model, the param load time is much longer than kpu calculate time!!! ok,let's forget realtime image application,but why the dvp can only input VGA image.
kpu is powerful ,consider it's price.but with so little ram and bad dvp , what can it do?