Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: >1 pipeline stage for L1D access?

Author: Tom Kerrigan

Date: 12:46:11 04/22/03

Go up one level in this thread


On April 22, 2003 at 03:36:33, Anthony Cozzie wrote:

>>This guy is a moron. He goes on and on about the P4's L1 dcache as if the
>>engineers at Intel all ride the short bus to school and somehow forgot to put
>>some more cache on the chip. Any idiot knows that the P4 has such a small dcache
>>because Intel wanted a cache with a 2 cycle latency instead of the 3 cycle
>>latency of the Athlon and P6. If Intel's measurements and simulations indicated
>>that a 3 cycle cache would be better, don't you think they would have put one
>>in? I mean, it's not like a 3 cycle cache is harder to design than a 2 cycle
>>cache (it's easier) and Intel's done it before, obviously. And if all Intel was
>>after was high clock speeds, putting low latency caches on their chips isn't the
>>way to go about it.
>
>While I agree with you (the P4 was designed for multimedia apps and not chess,
>and yes, it is good at them) I don't understand this talk about multi-cycle
>dcache latencies.  I was under the impression that *any* processor that took
>more than one cycle to access L1D was horribly slow, and my Athlon MP
>optimization guide says that the data cache access is peformed in one pipeline
>stage. (P219).
>
>anthony

It probably means that one stage is dedicated to cache access, but that doesn't
mean the access only takes one cycle.

http://www.xbitlabs.com/articles/cpu/display/athlon64_6.html

-Tom



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.