Subject: RE: [linux-audio-dev] The intel C compiler
From: Bob Colwell (bob.colwell_AT_attbi.com)
Date: Fri Feb 22 2002 - 00:46:33 EET
Hmmm, well then, let me give you my standard lecture on that subject.
Normalizing the clock rates of machines designed to achieve different
clock speeds is an odd exercise. There's a 400 MHz gap between Intel's
fastest CPU (Pentium 4 Northwood) and AMD's fastest. That gap exists,
not because AMD doesn't care about clock rate, but because the Pentium
4 was designed for 2.2GHz and the Athlon was not.
What makes this a slippery subject to discuss is that that higher clock
rate comes at a price: CPU efficiency. For the same reason that a race
car doesn't get the fuel economy of a compact car, a machine designed
to max out the circuits doesn't get the instructions-per-clock efficiency
of a machine designed for a slower clock. The faster clock demands more
pipeline stages, each stage implies latch/skew/jitter overhead, more
stages makes the branch mispredict penalty higher, and so on.
Normalizing the clocks is equivalent to pretending the slower-clock CPU
somehow found a way to achieve the higher clock without the attendant
overhead. Nice work if you can get it. But I don't know how to do that,
and I think if AMD did they'd have done it. Therefore, I believe you're
being a tad unfair to the Pentium 4 design.
If you'd like to compare Pentium III to Athlon on a clock-for-clock
basis, I think it's a lot more reasonable, and for that comparison I
think Athlon is a more efficient engine. Of course, the Pentium III
design is about 2 years older, so that's what you'd expect, no?
I promise to shut up now, since this seems pretty off topic even to
me. :-)
-BobC
bob.colwell_AT_attbi.com wrote:
>
> Saying the code runs faster than on an "equivalent"
> Intel processor requires that we agree on what
> equivalent means, a discussion that I suspect won't
> benefit this email list. (Equivalent process
> technologies? Equivalent clock rates? Equivalent
> aggregate performance? Equivalent product intro dates?) -
I meant same clock frequency... :)
- Jussi Laako
This archive was generated by hypermail 2b28 : Fri Feb 22 2002 - 00:39:22 EET