Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: move generation

Author: David Blackman

Date: 18:31:07 02/10/98

Go up one level in this thread


On February 10, 1998 at 15:13:08, David Fotland wrote:

>On February 10, 1998 at 09:22:13, Robert Hyatt wrote:
>>As I said, the *single* most important thing to do on modern micro-
>>processor architectures is to do everything possible to ensure cache
>>hits.  *every* miss is expensive.  Every hit is basically free.
[snip]
> Expect system early next century to waste more than 100
>instruction times on each cache miss.  Also expect the cache line
>size to increase, so more data is brought in with each miss.
>
>David Fotland

100 instruction times for a cache miss is fairly common right now,
especially
if there is a TLB miss at the same time, and there often is. There are
some
signs of hope that system designers are trying to improve latency, so
maybe
they won't get much worse than 100 instructions/miss for the next few
years.

Example: 200MHz pentium peaks at 2 instructions per clock or 2.5ns per
instruction, and typically 4 to 8 ns. Typical measured latency to main
memory
is 200ns to 800ns depending on how good the motherboard is, and whether
there
is a TLB miss at the same time.



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.