Building a 50 Teraflops AMD Vega Deep Learning Box for Under $3K
Carlos E. Perez
40413

Thx for article & ur point of view!
Do u try some more comprehensive DL experiments, cuz there could be some troubles with fp16 -> if you have deep NNs and moreover small lr you can encounter with NaN values in your NN after applying the grads to its weights (it’s true for tensorflow), since that you should be more accurate in your computations or perform some manual self-check to be sure that everything goes in the right way.

Show your support

Clapping shows how much you appreciated Denis Nemiroff’s story.