Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: HW based Crafty (Boule's thesis)

Author: Vincent Diepeveen

Date: 04:12:58 04/08/02

Go up one level in this thread


On April 07, 2002 at 22:39:33, Keith Evans wrote:

>On April 07, 2002 at 19:44:16, Tom Kerrigan wrote:
>
>>On April 07, 2002 at 12:50:36, Keith Evans wrote:
>>
>>>Do you have any idea what the spread is? Or what this is for a recent Crafty on
>>>an x86? Hsu mentioned 40,000 instructions per node as typical for a high-end
>>>program in his IEEE Micro article, but didn't explain how he arrived at the
>>>number.
>>
>>That is wronger than wrong. Are you sure he wasn't talking about how many
>>instructions it would take to implement DB's eval in software? Because even if
>>you assume an MPU that retires 2 instructions per cycle, that would result in a
>>program searching only 100k NPS on a 2GHz CPU. Programs were searching that fast
>>on CPUs from 7 years ago.
>>
>>-Tom
>
>I think that you are correct. Vincent says that Diep gets about 60k NPS on a
>fast K7, but it's definitely an outlier. I still wonder how Hsu got that number.
>Did they write a efficient software model of the DB eval?
>
>Regards,
>Keith

Yes, please take into your consideration that not all nodes are evaluation.
Only a small part of the nodes gets evaluated. This is true for all
programs.






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.