[linux-audio-dev] P3 vs K7

New Message Reply About this list Date view Thread view Subject view Author view Other groups

Subject: [linux-audio-dev] P3 vs K7
From: est_AT_hyperreal.org
Date: pe elo    27 1999 - 21:50:38 EDT


David Olofson discourseth:
>
> And AMD K7 beats a P-III hands down on the same rate, and is available at up to
> 800 MHz (cooled) right now. That 800 MHz box is nearly twice as fast as a P-III
> 550, which means... some 10 to 15 time what you've got in your box. And that's
> just ONE K7. They support SMP up to 8 or 16 CPUs... (Don't remember the
> figure.)

Interested parties should check out:
tomshardware.com/cpu/99q3/990823/
tomshardware.com/cpu/99q3/990809/

> And, then there's 3DNow! (SIMD - Single Instruction, Multiple Data, which
> means you can process 2 or 4 FP numbers in parallel), which boosts signal
> processing performance a lot. MMX enabled speed-ups between 50% and 400% by
> converting code from normal integer stuff, and 3DNow! should do at least half
> of that improvement for FP.

I find that my naive assembly language can often get a 3x speed-up
from 3dnow. More attention to instruction scheduling edges it closer
to 4. There are reasons that the speedup may not be as much in the k7
(i'm using a k6-2), but 2x isn't so bad. :)

> Oh, and AMD added a few special 3DNow! instructions
> suitable for signal processing. AFAIK, one is the multiply-and-accumulate
> instruction that used to be the reason why DSPs are a lot more efficient than
> normal CPUs for signal processing.

I'm afraid it's not there. There are, however, two new accumulation
modes, two float<->int32 instructions and a swapping move.

Eric


New Message Reply About this list Date view Thread view Subject view Author view Other groups

This archive was generated by hypermail 2b28 : pe maalis 10 2000 - 07:25:53 EST