Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: its possible to make 1 billion nodes/sec chip 2day- Hsu

Author: Uri Blass

Date: 12:31:54 10/17/02

Go up one level in this thread


On October 17, 2002 at 15:01:21, Robert Hyatt wrote:

>On October 17, 2002 at 14:23:19, Uri Blass wrote:
>
>>On October 17, 2002 at 11:58:48, Robert Hyatt wrote:
>>
>>>On October 17, 2002 at 11:17:43, Uri Blass wrote:
>>>
>>>>On October 17, 2002 at 10:23:59, Robert Hyatt wrote:
>>>>
>>>>>On October 17, 2002 at 03:34:44, Gian-Carlo Pascutto wrote:
>>>>>
>>>>>>On October 16, 2002 at 18:52:08, Robert Hyatt wrote:
>>>>>>
>>>>>>>You are simply relying on _one_ game.  What about 10 years of total domination
>>>>>>>in computer chess?
>>>>>>
>>>>>>10 years of domination ending 7 years ago means absolutely nothing now.
>>>>>>
>>>>>>--
>>>>>>GCP
>>>>>
>>>>>
>>>>>Only to those that don't "think".
>>>>>
>>>>>They dominated for ten years with hardware 100X slower than what they unveiled
>>>>>in 1997.
>>>>>
>>>>>If you take any reasonable program of today and make it 100x faster, do you not
>>>>>think it will
>>>>>totally dominate the other programs???
>>>>
>>>>If I take Fritz3(p90) and give it hardware that is 100 times faster then I
>>>>expect it to be weaker than Fritz7 of today inspite of a small hardware
>>>>advantage.
>>>>
>>>>The progress that was made in hardware+software from 1995 is more than being 100
>>>>times faster except fast time control.
>>>>
>>>>Uri
>>>
>>>
>>>If you put fritz3 on hardware 100 times _faster_ than fritz 7, I'll take fritz3
>>>in a match.
>>>
>>>This is _easy_ to test...  Start off 10x faster to see how bad 100x is going to
>>>end up.  10x
>>>will be murder
>>
>>I say 100 times faster than p90(the hardware that Fritz3 used to beat deep blue
>>prototype)
>>
>>If you put Fritz3 on 9000Mhz against Fritz7 on 2000Mhz when Fritz7 has bigger
>>hash tables I expect Fritz7 to win at least at long time control.
>>
>>Uri
>
>That isn't the point.  Thru 1994 deep thought dominated everybody, easily.  In
>1995 they
>lost one game to fritz.  In 1997 deep thought suddenly became 100x faster and a
>lot "smarter"
>with the new DB2 chips.

This is what Hsu claims but we have no evidence that it is the case.
First he had to admit that the 200M nodes that IBM claimed in the matches were
wrong and there were only 126M nodes so it is less than 100 times faster than 2M
nodes per second of Deep blue prototype and it make me suspect that there were
more lies.

Vincent claims that the effective number of nodes was smaller than 126M nodes
and 126M was not the number of nodes.

I do not claim to know.

About being smarter I am also not sure about it.
No doubt that deeper blue had more evaluation then deep blue prototype but
the evaluation was not tested enough and there were bugs.

Uri



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.