Computer Chess Club Archives


Search

Terms

Messages

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

Author: Vincent Diepeveen

Date: 08:02:26 04/09/02

Go up one level in this thread


On April 08, 2002 at 23:12:59, Robert Hyatt wrote:

>On April 08, 2002 at 07:02:21, Vincent Diepeveen wrote:
>
>>On April 06, 2002 at 22:09:03, Robert Hyatt wrote:
>>
>>>On April 06, 2002 at 00:53:39, Tom Kerrigan wrote:
>>>
>>>>On April 05, 2002 at 14:46:59, Robert Hyatt wrote:
>>>>
>>>>>Seems pretty obvious that "early deep blue hardware" meant something other
>>>>>than "deep blue hardware"...
>>>>
>>>>Yeah, except you said "original," not "early." Do you consider Chiptest
>>>>"original DB hardware"? Because I don't.
>>>
>>>I certainly consider "deep thought" to be "original deep blue hardware".
>>>
>>>As does most everyone, since that is where it _started_...
>>>
>>>>
>>>>>There _is_ no "precise number".  There were three complete revisions of the
>>>>>chess processor.  I haven't seen anything that said all three had the same
>>>>>number of cycles in each operation or that they didn't...
>>>>
>>>>So in other words, you don't know the numbers (because if you did, you would
>>>>know whether or not they were the same). So your information-free replies
>>>>continue to mistify me.
>>>>
>>>>>I think you pointed out the flaw yourself.  2000 instructions at 2ghz is not
>>>>>_nearly_ enough to do a node.  And a 12mhz FPGA is a very slow FPGA.  100mhz
>>>>>is more like it for SOTA...  I'll take on that 2ghz general-purpose CPU any
>>>>>time you want...
>>>>
>>>>First, my own program would search more than 1M NPS on a 2GHz chip. Which means
>>>>fewer than 2k cycles per node. Which means ~2k instructions per node, and
>>>>possibly less. Which means that not only are 2k instructions "nearly" enough to
>>>>do a node, they ARE enough to do a node.
>>
>>>I believe I said a "real chess program"...  I don't know of any "real" engines
>>>that search 2K instructions per node...  I'm also talking about _real_ nodes...
>>>Just to be clear...
>>
>>Fritz3 was under 1000 clocks a node at old P5 hardware which issued
>>at most 2 instructions a clock. Now Frans is very good in programming
>>such that he gets close to that, but definitely it's less than 2000
>>instructions a clock.
>
>Aren't you talking about the _old_ Fritz using root-preprocessing and
>piece/square tables?
>
>
>>
>>Fritz3 when upgrading it to current hardware and improving hashtable
>>a bit would search easily 20 ply or something in middlegame at tournament
>>level.
>
>Did Fritz 3 use forward pruning?  without that it has no chance to hit 20
>plies.

Fritz3 was very near to material counting program. I don't know
whether he used forward pruning. perhaps last few plies to replace
nullmove a little, but at 20 ply depth that's just 3 last plies at
most.

As experiments lately have indicated getting 20 ply with a near to
material only evaluation is way easier than getting it with a complex
evaluation, simply because everything gives a cutoff for nullmove
and even stupid moves like 1.a3 might be a killer move for white
whereas in modern chessprograms 1.e4 or 1.d4 is considered better,
which means that there are more complications at the board where
nullmove fails.

In short, it seems to me that fritz7 uses way more forward pruning than
fritz3 ever did.

Best reards,
Vincent

>
>>
>>>
>>>
>>>>
>>>>Second, what the hell are you talking about with regard to FPGA speeds? "A 12MHz
>>>>FPGA is a very slow FPGA"? It's easy to come up with some logic that would run
>>>>at less than 1MHz on the fastest FPGA ever. Your apparent notion that FPGA clock
>>>>speed is somehow independent of the design that's loaded into the FPGA speaks
>>>>volumes about your ignorance of what an FPGA actually is.
>>>>
>>>>-Tom
>>>
>>>
>>>If that is as _fast_ as the specific FPGA you want to use can be clocked,
>>>then _yes_ it is "very slow".
>>>
>>>Nothing more to say...  There are parts available for a year or more that
>>>run over 75mhz...  A FPGA certainly has a max clock speed regardless of _what_
>>>is "loaded into it".  This clock speed might be significantly lower due to the
>>>thing being "loaded" of course.  But there _is_ a max no matter what is loaded,
>>>and _that_ is the raw speed number I was referencing..  Everything has a max
>>>due to various things from gate delays to whatever you want..



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.