Originally Posted by Seronx
The increase isn't totally IPC, some of it is enhanced Word Level Paralleism.
128-bit = 4x32-bit ops or 2x64-bit ops
256-bit = 8x32-bit ops or 4x64-bit ops
There is also Skylake having 3x256-bit Packed Arithmetic Integer units. Compared to (Piledirvers)Excavators (3)2x128-bit Packed Arithmetic Integer units. XV to ZN returns to Piledriver levels for Packed Arithmetic Integer @ 3x128-bit.
This may technically be true--the Skylake part may be packing more work into fewer instructions--but the difference between instructions per clock and performance per clock is hard to distinguish, and ultimately, academic.
Core for core, clock for clock, Skylake is way
, faster than Excavator. So much so that I'm surprised the ~80% figure strikes anyone as being particularly outlandish.
Originally Posted by FattysGoneWild
They down clocked the Intel cpu to go against Zen? Really AMD? REALLY?!?!?! "We are back" No you are not if you have to down clock a competitors chip to match the best of what you are showing.
Can AMD for once in their life be 1 step ahead of Intel? I mean just ONCE. I don't care if it never happened again. Just show you are capable. They seem quite content on always finishing last with their cpu and gpu offerings.
They were illustrating performance per clock, not trying to pit what is probably early silicon against mature production parts in a clock speed contest.