2 ms·
Sorry, should have written "compute units" - the point is as you point out that flops is one thing but branch prediction and feeding those floating point units
by alkonaut 11y ago
Sorry, should have written "compute units" - the point is as you point out that flops is one thing but branch prediction and feeding those floating point units is another.
There has been a surge in the tractability of massive but simple linear algebra problems lately, such as deep learning, which might have given the impression that GPUs are the answer to any supercomputing.