So while looking into the possibility that an MMX at 166 is obnoxiously over-fast if I want to represent a plain Jane P166, I discovered Intel pulled a bit of a sneaky with the iComp 2.0 rating... I mean they documented it, but who all gonna dig that out and not just glance at the gee whiz table to see P166MMX looking a large portion of double the speed of a P166.
But hol' up there frendo, what's this. The P166 got a 430FX board and EDO RAM and the P166MMX got a 430VX with SDRAM, shenanigans! I think it wasn't as bad as the FX having crappy old asynch and being 30% plus behind, but it might still have been just into a double digit percentage, vs if it was also allowed to run at full tilt on the VX with SDRAM. Then of course the new MMX testing got weighted strong-ish, and some tests seemed to favor the doubled up L1 of the MMX. Anyway, looking at game and synthetic benches on Thandor.net, they seemed pretty close on most stuff, with sometimes the MMX running ahead.
Anyway, seemed to end up with the impression that it's only MMX supporting stuff and stuff with core routines between 16 and 32kb which will really "run away" on the MMX, so I could use it and not feel toooooo dishonest. So I think it's a case of "If modding the reg to 3.3V is easy to figure out and do, do it, if it isn't, shrug and leave the MMX in" would be testing with MMX until satisfied board is "as delivered" functional before messing.
I thought I found a page some years back where someone did iComp 1.0 ratings for later intel CPU, it doesn't go past the Pentium Classics officially. But damned if I can dig it out now.
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.
