SYN said --
"6-8x performance increase when compared to a MP800 with a simple recompile is not possible with any chip that is even remotely close to being ready for mass production... Has this ever happened? even the 604e to G3 did not provide such an increase.
The only case in which this could in fact be true is if the App was heavily reliant on AltiVec... or completely starved of memory bandwidth on current machines...
6-8x is quite a speed jump."
On the contrary, I think this jump would be (barely) possible, even leaving AltiVec out of it, with what I expect to be the next machines. Assume that the app fahre451 is talking about runs without MP tasks on a dual-800, and that the sealed-up machine that his new tests were run on was one of the 2.4-GHz G5 models that Architosh.com and other sites mention, then we have SPEC2000 (G4 DP-800) == 480 and SPEC2000 (G5 SP-2400) == 2046, so that the speedup would be 2046/480, or 4.3 x, quite consonant with the subjective "6 to 8 times" reported. See my CPU-characteristics chart at:
On the other hand, the sites and figures that I am relying on are not universally accepted (to say the least). Witness these responses to the Register/MacOSRumors/Geek.com/Architosh.com
numbers (Oct - Nov 2001):
obvious fakes (Oct 20 2001).
Those numbers are obviously fake. I predict that the G5 will actually score closer to 50% of those faked scores. The Register got its numbers from MacOSRumors, a site notorious for bogus and wildly over-optimistic performance claims (a site that makes Apple's bogus performance claims actually look conservative).
These SPECS are bullshit. Not because I can't have the G5 to be that fast but because such a score would rely on a groundbreaking design. The G5 isn't. It's a nice design, very nice, but it's still human -)
The specs are greatly inflated (Nov 07 2001).
I'm telling you people that the specs are way too good. There is no way the integer and floating point performance is 3X as good as a P4 at the same clock speed. You're all going to be in for a big disappointment when the G5 is finally released and tested. - by Mark
But I believe that, based on the public information from Motorola and the "moles" used by the rumor sites, which over the last three years, as far as the G5 goes, have all shown an astonishing degree of consistency and plausibility, the referred-to specs are not fakes.
To the other issue -- re-compiles or re-writes being necessary for a new chip -- Apple and Motorola have assured us that no recompile will be necessary to run G4 binaries on the G5 (when it comes), though a simple re-compile can greatly speed up 64-bit operations. (Does fahre451's code do 64-bit arithmetic?)