Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Opinions? A Crafty experiment...

Author: Robert Hyatt

Date: 17:31:48 05/24/04

Go up one level in this thread


On May 24, 2004 at 17:21:52, Vincent Diepeveen wrote:

>On May 24, 2004 at 15:04:25, Robert Hyatt wrote:
>
>>On May 24, 2004 at 14:48:30, Vincent Diepeveen wrote:
>>
>>>On May 24, 2004 at 14:37:06, Slater Wold wrote:
>>>
>>>>On May 24, 2004 at 14:32:38, Vincent Diepeveen wrote:
>>>>
>>>>>On May 24, 2004 at 14:09:56, Robert Hyatt wrote:
>>>>>
>>>>>>On May 24, 2004 at 12:56:56, Slater Wold wrote:
>>>>>>
>>>>>>>On May 24, 2004 at 12:29:04, Russell Reagan wrote:
>>>>>>>
>>>>>>>>On May 24, 2004 at 12:06:46, Slater Wold wrote:
>>>>>>>>
>>>>>>>>>In a thread below, there is a post about Blargh, ICCs top computer.  It has won
>>>>>>>>>60% of its games at standard time controls, and only lost 16% in almost 3,000
>>>>>>>>>games!
>>>>>>>>>
>>>>>>>>>I am interested to know the OoO (opinion of outcome) of EVERYONE at the
>>>>>>>>>following experiment, I will be in contact with the operator of Blargh, and will
>>>>>>>>>try to set this up.
>>>>>>>>>
>>>>>>>>>Crafty 19.13 on an AMD FX53 (64-bit) vs Shredder 8 on a Dual Opteron 2.2Ghz
>>>>>>>>>(32-bit)
>>>>>>>>>
>>>>>>>>>10 game match.  120/0.
>>>>>>>>>
>>>>>>>>>Opinions?
>>>>>>>>
>>>>>>>>Crafty on equal hardware probably doesn't stand a realistic chance against
>>>>>>>>Shredder. Only Fritz scores at least 45% against Shredder on the latest SSDF.
>>>>>>>>Junior and Hiarcs score at least 40%. Crafty would be behind those engines on
>>>>>>>>equal hardware by a non-trivial margin (take the Junior-Crafty 10x experiment
>>>>>>>>for example).
>>>>>>>
>>>>>>>OoO please...
>>>>>>>
>>>>>>>>What is interesting about a Shredder-Crafty match where Crafty is at a hardware
>>>>>>>>_disadvantage_?
>>>>>>>>
>>>>>>>>BTW, that really explains a lot about why he is #1 on ICC. You have a freakin'
>>>>>>>>FX53 and you're at a hardware disadvantage! Wow...
>>>>>>>
>>>>>>>It's a disadvantage, but a small one.  How small?
>>>>>>
>>>>>>It is at least 2x disadvantage.
>>>>>>
>>>>>>You did notice that box is a _dual_???
>>>>>>
>>>>>>Shredder has an SMP version which I would assume he is using...
>>>>>
>>>>>Comeon you need to be NUMA to run well on a dual opteron in 32 bits. Shredder is
>>>>>SMP and has not released a 64 bits executable yet for xp64.
>>>>
>>>>A point not everyone is getting, obviously...
>>>
>>>Yes, shredder is missing a factor 3 somewhere thanks to this.
>>>
>>>It's faster single cpu A64 in 64 bits than dual Xeon 3.2Ghz
>>>
>>>single cpu net2003 default compile 32 bits, latest diep version 45% faster at
>>>2.2Ghz opteron than 2.127Ghz K7.
>>>
>>>So single cpu version 135k nps (both numa & smp version)
>>>NUMA version scales factor 2.0, SMP version from diep gets 180k nps
>>>at dual opteron 2.2Ghz. This all in 32 bits mode. SMP version might suffer
>>>though from fact that dual machine has not all banks filled with RAM yet.
>>>
>>>I hope later today to have a go at 64 bits XP (if i can get it to work there so
>>>soon, no clue how to lock in 64 bits XP).
>>>
>>>Shredder and DIEP, in contradiction to crafty, do a lot of RAM lookups.
>>>
>>>So NUMA versus SMP is a bigger penalty.
>>>
>>>Based upon hard facts extrapolation for shredder at dual opteron in 32 bits
>>>means : 1.5 * 1.45 = 2.175 times faster in nps.
>>>
>>>Crafty 32 bits executable (version 19.3 was only one i found so quick at his
>>>ftp) at bob's ftp gets 770k nps at 2.2Ghz opteron.
>>
>>Please do not make up numbers when actual 2.2ghz opteron log files are available
>>on my ftp box.  Crafty gets 2M on a 2.2ghz opteron, period...
>
>Windows executable at your ftp in 32 bits gets 770k. period.

Strange nobody _else_ gets such an incredibly slow number.  Could it be that
once again your imagination is over-active?  Can someone _else_ (slate has an
FX53 I think) run the 32 bit executable to see if it is _really_ that slow?
That 64 bit speeds me up by more than 2x, something you have _always_ said will
never happen.  You have to pick a position and stick with it, otherwise you get
caught (again) on _both_ sides of the fence, which is physically impossible...
At least for _most_ of us.

Any reliable testers want to try this????



>
>>
>>>
>>>Is it in nps more than 2 times faster at opteron in 64 bits than this?
>>
>>See above and check the log files for yourself.  And 32 bit crafty doesn't do
>>770K nodes per second.  That is slower than a single 2.4ghz xeon, much less my
>>2.8's and beyond...  that's a bogus number.
>
>Then you basically say that your ftp site has worthless executables.

No.  I "basically say" that you are unqualified to test a mouse-trap, much less
do any unbiased testing on _my_ program.  You _do_ remember your "your program
doesn't produce _any_ speedup at my dual K7" don't you?  Utter nonsense as
several others have proven...


>
>Oh by the way it doesn't compile SMP under windows at all.
>
>Seems i'm one of the few who now and then compile your achillesheel....
>


I only have two "achilles heels".  Both real close to my ankles...





>>
>>>
>>>If so then having a native 64 bits compile for opteron has more effect than a
>>>dual opteron in 32 bits mode for that exe.
>>>
>>>>>Equal hardware SSDF shredder8-crafty : 43.5 - 3.5
>>>>
>>>>I was going to look that up later, thanks!
>>>>
>>>>So we can expect in a 10 game match, nothing better than 9 - 1 from Crafty,
>>>>based on SSDF results.  Interesting...
>>>
>>>more like 10-0
>>>
>>>Just use a book that is not committing suicide for shredder.
>>>
>>>Faster hardware shredder extrapolates its rating better than crafty.
>>>
>>>Of course do not forget to use book.bin for crafty and no commercial books.



This page took 0.01 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.