Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Hydra Mystery Remains Unsolved

Author: Robert Hyatt

Date: 17:30:13 02/17/04

Go up one level in this thread


On February 17, 2004 at 15:23:57, Bob Durrett wrote:

>On February 17, 2004 at 13:01:24, Robert Hyatt wrote:
>
>>On February 17, 2004 at 12:25:18, Anson T J wrote:
>>
>>>On February 17, 2004 at 10:30:32, Robert Hyatt wrote:
>>>
>>>>On February 17, 2004 at 08:41:16, Bob Durrett wrote:
>>>>
>>>>>
>>>>>
>>>>>The fact is that Hydra whipped a bunch of conventional chess computers at
>>>>>Paderborn.  That fact is indisputable.
>>>>>
>>>>>How???
>>>>>
>>>>>How could Hydra, chugging away at the clock rate of a slow snail, win against
>>>>>the high-nps conventional machines?
>>>>
>>>>
>>>>I don't understand the question.  Hydra probably hit speeds of 15-20M nodes per
>>>>second.  How is that "a slow snail"???
>>>>
>>>>It was the fastest thing playing there by a factor of at least 4x...
>>>
>>>I think he is talking about the clock speed of the boards. I don't know the
>>>clock speed of the boards but I would imagine they are slower than GHz.
>>
>>Yes, but who cares?  It is the NPS that determines how fast a chess program
>>searches, and their NPS was above anything else by a big margin.
>
>Bob, please indulge a "slow learner."  I still don't get it.  Are you saying
>that the best way to get ***really*** high nps rates is with hardware [maybe
>such as used by Hydra?] as opposed to using a PC?

The idea is this:  Special-purpose hardware will _always_ be faster than a
general-purpose machine.  It will also cost more, because it is special-purpose
rather than using generally-available off-the-shelf parts.  This is true of most
anything you care to do.  If you design for a specific task, the result is more
efficient than taking a readily-available "solution" and applying it to a task
where compromises are required to make things work...

>
>Incidentally, I am really feeling ignorant right now.  How did Hydra get such
>high nps?

Special purpose hardware.  Very much like belle/deep-thought/deep-blue.  DB
required 10 hardware clock cycles per node.  Running each chess chip at 20-24mhz
produced 2-2.4M nodes per second per chip.  Apparently the FPGA used in Hydra is
running at 30mhz, and apparently they chose a design similar to
belle/deep-thought/etc since the 10 clock per node speed seems to hold...


>
>I hope you don't mind helping a beginner along on this confusing stuff. [Mark
>thinks I'm pretty dumb.]
>
>Bob D.
>
>>
>>Would it matter if someone showed up with an Indy car that ran 300mph but with
>>an engine that only turned 4,800 RPM?  Would that be considered "slow" since the
>>rest are hitting 10K+ on their tachs???
>>
>>
>>>>
>>>>
>>>>
>>>>>
>>>>>They say "nps isn't everything."  But could the truth be "nps isn't anything"?
>>>>>
>>>>>Maybe conventional wisdom ["The Earth is flat"] isn't right after all.
>>>>>
>>>>>Does anybody understand what happened?  I feel that the results were monumental!
>>>>>
>>>>>Bob D.



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.