Re: RC5 not the only place it is fast
Originally posted by gopher
Photoshop, and Genentech Blast. And if you want to go further
the folks at Yellow Dog Linux http://www.yellowdoglinux/ have made Black Lab Linux which is Altivec optimized.
Wooptie-doo!
So before you get to saying the G4 1 Ghz is not faster than the Pentium IV 2.4 Ghz, there are many other applications that use Altivec where it is faster. And it is those applications which require it, are where it counts.
Agreed. But generally, the G4 is much slower than pretty much all of its competition.
The Mhz myth is a myth, and we should ask how many floating point calculations a second the Pentium and AMD can do. The G4 dual 1 Ghz machine can do 15 billion floating point calculations a second.
That's according to Apple marketing, with perfectly tuned code under perfectly controlled conditions. It's also 15 billion
single-precision, not double-precision, FLOPS. If you want to listen to Apple marketing, that's fine, but if you want to listen to reality, here's a little dose of that for ya:
SPEC_CPU2000 - the most widely accepted and accurate CPU benchmark currently in existence:
1GHz PowerPC G4: 306 int / 178 fp (peak, per CPU)
1.13GHz Pentium III: 461 int / 320 fp (base - peak is higher)
2.2GHz Pentium 4: 790 int / 779 fp (base - peak is higher)
Why, the 1GHz G4 can't even hold its own against the Intel Sh*tanium:
800MHz Itanium: 358 int / 715 fp (base - peak is higher)
Of course AltiVec will make those results less embarrassing, but ONLY with single-precision floating point code. So the G4 excels at a few specific, limited tasks - it really does suck at most everything else.
Now to get those developers to learn how to use Altivec to their best advantage. They focus too much on graphics cards and not on the processors.
Why should they? Where's the market? They've optimized Photoshop and FCP and whatever other
niche products, but where is their incentive to completely re-write and fork the fp-intensive parts of their code to make them faster on the Mac but incompatible with x86, when they can write that code once and have it run almost as fast as it possibly could on what comprises 95%+ of the desktop market without any special optimization whatsoever. This is why I say PPC needs a robust and capable FPU in the place of, or alongside, AltiVec.
Alex