Computer Chess Club Archives




Subject: Re: perft records

Author: Sune Fischer

Date: 04:37:35 09/07/04

Go up one level in this thread

On September 07, 2004 at 07:13:43, Reinhard Scharnagl wrote:

>On September 07, 2004 at 04:50:35, Sune Fischer wrote:
>>On September 06, 2004 at 07:43:51, Peter Fendrich wrote:
>>>What free programs have the fastest perft and what are the figures?
>>>Please, If you give figures also add processor, compiler and environment!
>>Frenzee does perft 7 on the initial position in 28 seconds on my 2 GHz Athlon.
>>However that is using hash and "the Uri trick" with no make/unmake at the last
>>>I want to compare with a new concept that isn't coded yet...
>>Just realize that this is really comparing apples to oranges.
>>I tried using the above fully legal generator in the program as
>>opposed to the old incremental generator, but all I saw was a big slowdown.
>>I don't recall the exact number but I believe it was around 10-15% slower than
>>the incremental approach.
>>My conclusion is that at least for the frontier nodes a fully legal move
>>generation is not worth it, at least not for a bitboarder that can easily do a
>>much faster incremental pseudo generation.
>Hi Sune,
>last statement could been proved easyly by perft or it is simply an opinion.

Hmm, how do you test this in perft?

A long move list takes longer to generate and it takes longer to sort it.
That would be double penalty in a real program but in perft there is no overhead
in working on a long movelist since you'll be needing all the moves anyway.

I don't see how to "fix" this in perft to make it a good test, do you randomly
start throwing moves away to simulate beta cuts? :)

I think it is far better to implement both and test it in a real program, that's
a bit extra work perhaps but it's also a no nonsense direct test.

>My opinion is, that bitboarders neither are really faster nor slower.

Yeah hard to say, it's not easy to construct a valid test.


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.