Funny how this turned around - until very recently low-precision (<32 bit floats) arithmetic was intentionally limited on consumer cards - iirc FP16 rate was li

Funny how this turned around - until very recently low-precision (<32 bit floats) arithmetic was intentionally limited on consumer cards - iirc FP16 rate was limited to 1/100th or something crazy like that. Intel being the true pioneers they are not only made this available at full performance to their customers, but even the default!