Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: a number and two questions for bob

Author: Robert Hyatt

Date: 13:43:23 05/05/04

Go up one level in this thread


On May 05, 2004 at 15:55:51, Gian-Carlo Pascutto wrote:

>On May 05, 2004 at 15:43:06, Robert Hyatt wrote:
>
>>So how do you answer the question I was asked?  "What kind of speedup do you get
>>in real games?"  Not "what kind of speedup do you get on test positions?"
>
>I don't think the concept 'speedup in a game' makes sense. You can
>get a speedup on positions from a game, but that's not the same thing.

That is _exactly_ what I computed.  I then computed the overall (average)
speedup as well...  by averaging the individual positions...  Again, it seems
that you didn't read the paper as this was all explained in great detail...


>
>>My test positions were run in the normal way, but with a correctly pre-loaded
>>hash table at the start.
>
>Yes - thus keeping the headstart and making the measurements dependant.

There was _no_ headstart.

Since you didn't read the paper, apparently, think about this>

the 2-cpu game log shows that the last PV came out at exactly 2:00 into the
search, but the search continued until exactly 4:00 where time ran out.

I now did the 1-cpu test and noticed that the same PV comes out at 3:30.  I get
a speedup of 210 / 120 = 1.75.  I now go back with the 1 cpu test, and run the
same position to a time limit of 1.75 * 4:00, which is about as close as I can
come to letting the 1-cpu test search the _same_ sized tree as the 2-cpu test.

So there was no "head-start" whatsoever...  All this was explained...

After the second search to the longer time limit, I then have the hash table set
for the _next_ position... with the 1-cpu and 2-cpu hash being nearly identical.
 I didn't think but it would have saved a lot of time to re-run the 16 cpu test,
and after each search write the hash to disk.  Then re-load it correctly for
each other test...  I certainly had enough disk space..  But my way gave the
same effect...

>
>--
>GCP



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.