3 ms·
Why do they call it that then? I never really looked into it that much and just took as a measure of a certain type of compute capability ( FP16 or FP64 right?
by Melatonic 4y ago
Why do they call it that then? I never really looked into it that much and just took as a measure of a certain type of compute capability ( FP16 or FP64 right? )
- wmf 4y agoThe SIMT architecture makes it look to the programmer like each FPU is a separate core, but all the cores in an SM have to run in lockstep to get good performance.
- monocasa 4y agoThey use that metric because it makes their marketing specs look nicer. Ultimately it's a count of the number of SIMD lanes.