@RickMerrit: thanks for the writeup, didn't see you at the Computer History Museum for the 40th Anniversary of Spice event (most ISSCC attendees made it there!).
I wonder why IBM isn't using their Power7 platform to push the clock speeds instead of the z196. The Power6 older generation was already clocking at 4.5GHz whereas Power7 had already reached 5.0GHz (8-core version?). The Power7 also has a L3 cache of 32Mb max.
I strongly suspect Rick is right. Out of order techniques don't yield 40% perf improvements across the board, all by themselves. 20 - 25%, maybe. How well the compiler understands the new, complex microarch is another strong factor. After that you have cache design, bus design, policies on TLB usage, provisions for coherence, and how aggressive you were on lots of other things.
Actually it is good to see an edging up on the frequency. The move to an out-of-order design (40% improvement) is a key advancement along with an increase of 18% in frequency.
I would expect the frequency might edge up a little more over time with the same basic design, but the "step" function of 40% for OOE is the kicker..
I actually know something about both projects, ex-IBMer and ex-MIPS executive..
Blog Doing Math in FPGAs Tom Burke 15 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...