Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How will Junior Fare at Dortmund?

Author: Robert Hyatt

Date: 10:11:59 06/02/00

Go up one level in this thread


On June 02, 2000 at 10:52:20, Pete Galati wrote:

>On June 02, 2000 at 10:42:17, Robert Hyatt wrote:
>
>>On June 02, 2000 at 10:13:35, Albert Silver wrote:
>>
>>>On June 02, 2000 at 02:01:29, Robert Hyatt wrote:
>>>
>>>>On June 02, 2000 at 01:44:29, Pete Galati wrote:
>>>>
>>>>>On June 02, 2000 at 00:58:00, blass uri wrote:
>>>>>
>>>>>>On June 02, 2000 at 00:18:23, Pete Galati wrote:
>>>>>>
>>>>>>>On June 02, 2000 at 00:06:40, Rick Luce wrote:
>>>>>>>
>>>>>>>>Junior 6 is slated to participate in the Dortmund tournament in July.  How well
>>>>>>>>do you think it will do?  My guess is that Junior will fare poorly against such
>>>>>>>>a strong field and could easily finish last.
>>>>>>>>
>>>>>>>>By my calculation, Fritz performed at about 2590 level at the Dutch
>>>>>>>>Championship, if you discount the two games which were essentially forfeit wins.
>>>>>>>> If Junior performs at a similar level, it could be in trouble.  The competition
>>>>>>>>at Dortmund will be much stronger including the likes of Khalifman, Anand,
>>>>>>>>Huebner, Leko etc.
>>>>>>>>
>>>>>>>>Also, has there been any indication that any of the GMs will refuse to play
>>>>>>>>Junior?
>>>>>>>>
>>>>>>>>-Rick
>>>>>>>
>>>>>>>Do you have a link for a site with info about the tournament?  Which Junior?
>>>>>>>What hrdware will it run on.  My unqualified opinion, not knowing any answers to
>>>>>>>those questions anyhow, is that Junior will do better than Fritz did.  Just a
>>>>>>>gut feeling.
>>>>>>>
>>>>>>>Pete
>>>>>>
>>>>>>I remember that Junior will use 8 processors so it will use better hardware than
>>>>>>the hardware of Fritz.
>>>>>>
>>>>>>Uri
>>>>>
>>>>>Thanks, yes, I found this at the Chessbase site; "in the Super-GM tournament it
>>>>>will have an eight-processor system".  So Junior should do very well.
>>>>>
>>>>>Pete
>>>>
>>>>
>>>>Don't panic here.  8 cpus is not twice as fast as 4 cpus, in the world of SMP
>>>>searching.
>>>
>>>Yes, meaning it most likely won't even get an extra ply as opposed to 4 cpus
>>>unless they are also using newer and faster processors (like the new 700 Mhz
>>>Xeons) for the occasion. All the same, the matchups are fascinating. It's hard
>>>to say how well it will do but if Xie Jun's games are any indication, Junior may
>>>still be in for some serious trouble. Barring some of the absurdities that took
>>>place, like the operator mistakes, the book problems, etc... Xie Jun did very
>>>well against it, and I expect that these players will most likely make mincemeat
>>>of it. Among other things, I think that contrary to the Dutch championship, all
>>>the players here will be well-prepared against Junior as no one will want to
>>>become its first Super GM scalp. Still, Junior's resilience should help save it
>>>from many difficult positions so I predict a TPR slightly over 2600.
>>>
>>>                                       Albert Silver
>>
>>
>>The new 700 xeons won't go 4-way.  To date, the fastest 4-way boxes are based on
>>the 550mhz xeons that are about a year old now.  We have been waiting on faster
>>4-way chips, but Intel has not delivered.
>
>If this accounted for a major share of their market, you know they'd deliver.
>
>>The 8-way boxes are huge kludges
>>that even in their best cases don't go 2x faster, just because memory can't keep
>>up.  And then there is SMP search inefficiency.  I ran on an 8-way box and was
>>not impressed at all (I ran on a 16-way alpha and was impressed out of my mind,
>>however.  :)  )
>
>Are there no advantages to the 8cpu boxes in this situation?
>
>Pete

They are faster, but not a lot faster.  IE I ran a common benchmark, which is
simply a compute-bound program that requires a good bit of memory bandwidth to
run.  On my quad, if you run one copy it runs in N seconds.  If you run 2
copies, it takes 2N+M seconds where M is not really too big and represents a
memory bottleneck.  I lose about 7% per cpu doing this...  On the 8-way box,
this benchmark would nor run more than 6 times faster, while on my 4-way box,
it runs about 3.8 times faster.

The 8-way has a memory bottleneck.  The 4-way interleaving in the 4-way and
8-way boxes is barely enough for a 4-way machine.  8-way finds memory accesses
a real delaying problem.



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.