Author: Robert Hyatt
Date: 17:41:26 03/05/00
Go up one level in this thread
On March 05, 2000 at 15:44:41, Tom Kerrigan wrote: >On March 05, 2000 at 14:38:16, Eugene Nalimov wrote: > >>You forget that P6/PII/PIII can run at much faster frequence - exactly because >>of a longer pipeline (and P5 is no exception here, please look at the PPC - it >>also have a very short pipeline, so it cannot run fast enough to remain >>competive with Intel). So when you are comparing them at the same frequence you >>are not doing the right thing. > >Yeah, it's an oversimplification of the issue. There are other things to >consider, too, like uops vs. x86 ops, and the P6's "extra" load/store units. I >think the argument got pretty far off-topic. > >But I still think that saying the P6's pipes are more full than the P5's pipes >without proof is pretty ballsy, for the reason I explained. > >-Tom The proof is intuitively obvious to the casual observer, and only takes a bit of reasoning to expose: Why do you suppose Intel dumped the p5 core, and burned all those transistors making the p6 core, if they weren't convinced, beyond a shadow of a doubt, that the p6 core was better for the general case? The answer is they wouldn't have don't it. They run zillions of simulations to predict peformance before they commit something to silicon. An experienced assembly programmer can look at how the two machines operate and intuit that the p6 is better...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.