Silverforce11
Lifer
- Feb 19, 2009
- 10,457
- 10
- 76
Integer FMA
That's it. I'm pretty sure the only reason Nvidia doesn't have it is because of back-room deals between them, since AMD has had it since 5xxx series, and Nvidia hasn't had it on anything.
Maybe Maxwell will have it. No clue.
Only useful when working on integers though, which really shouldn't be most of the real GPGPU work anyways.
You know NV market gk104 based and dual gk104 Teslas for the specific purpose of high sp integer HPC work (oil discovery simulations, mapping)? Thats why its expensive even-though it is poor in "real" HPC work...
Point is, HPC work all differ in their demands. But NV have that market due to a massive head start (good leadership) and wide usage of CUDA.
