4 ms·
If all you need is reduced precision, then it’s fair to compare the two. I’d also assume memory bandwidth matters just as much as TFLOPS for ml workloads.
by cobookman 9y ago
If all you need is reduced precision, then it’s fair to compare the two. I’d also assume memory bandwidth matters just as much as TFLOPS for ml workloads.