Hi, Theano support float16 with the new back-end. This is working, but we still have a few cases where we have upcast to float32 or don't apply numerical stability optimization. We are working on that.
To clarify, we have storage float16, when the above is done, we will add computation in float16. But don't forget only the P100 GPU have quick float16 computation. All others Pascal GPU I have see (except maybe embeded GPU, I didn't check those) have minimal and slower float16 computation then float32. Fred On Tue, Oct 25, 2016 at 5:57 PM, Isaac Gerg <[email protected]> wrote: > Does theano support fp16 for 6.x compute cards? > > -- > > --- > You received this message because you are subscribed to the Google Groups > "theano-users" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
