Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Fritz5 and memory

Author: Don Dailey

Date: 13:56:06 05/29/98

Go up one level in this thread


On May 28, 1998 at 16:02:52, Georg Langrath wrote:

>I have just bought Fritz5, and I am very satisfied with it. But there is
>a disadvantage with it that everybody that want to buy it should be
>aware of. It is enormously memoryhungry. In the manual they recommend 72
>MB! memory for a Pentium 200 in average 3-minutes play. And it is so.
>When the beast has eaten the hashmemory it nearly stops analyzing. To
>analyze longer time for example 15 minutes is impossible for usual
>homecomputers.
>I myself always play on shorter time, so my memory is enough (32MB on
>Pentum133). But if I should like to play tournament level I had to
>upgrade. The formula according to manual is 2 x HZ x minutes.
>
>Georg

I've read all the responses to this post and am surprised by some of
them.   I think I will try to clarify some of the myths.

1.  Bigger hash tables sometimes makes the program slower.

There are only 2 reasons this happens sometimes.  Extra memory can
reduce cache hits a little, not a very large effect.  The seond thing
that can happen is the search will change a little with any hash table
size change.  This is quite normal.  The effect will occasionally and
randomly make a bigger hash table look at more nodes.  But ON THE
AVERAGE
having bigger hash tables will reduce the node counts.   To test these
things do not run 1 position, but run several and average them together.


2. At some point bigger hash tables don't seem to help.

This is true, if you don't run deeper searches.  If the hash tables
do not get filled up, making them bigger is unlikely to help much
if any (but it won't hurt either.)  Bigger hash tables (on the average)
will ALWAYS help if you run a position long enough.


3.  Fritz stops searching if the hash tables are too small.

This is completely ridiculous.  What we see is that Fritz saturates
its hash tables more quickly than most if not all the other chess
programs.  When this happens there is a larger multiplier going
from one iteration to the next.  This happens for all chess programs,
but Fritz is so fast it gets to this point sooner.   This should be
considered a GREAT thing, not a BAD thing.  It's just more proof that
Fritz is faster than the other chess programs.

From a human subjective point of view, all programs seem to bog down
when they've been thinking a minute or so.  This is mainly just our
own perception or impatience.  We think, gee it got up to 10 ply in
just a few seconds or so but it's taking forever to get up to 12.
This effect may seem bigger on Fritz which gets up there really
quickly.

- Don






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.