[Beowulf] Re: Q&A: BeOS and IBM Cell processor

[Vincent Diepeveen]

Actually Douglas there is in games like chess really a lot of data.

Be welcome to enter statistical paradise, you can walk around for years 
there
and still will be able to fill your computer with statistics :)

There is a lot of data of all kind of compilers, and it differs from program 
to program.

Especially visual c++ 2005 which most are using now.

Some of authors post here:
   www.talkchess.com

And go to the computerchessforum.
Happy i'm not the one doing a 60% claim, others are though.

So just please ask there. In fact some of the commercial guys are posting 
there too.

Most aren't posting at rec.games.chess.computer

A good person to ask is the Bodo programmer name: Joel Veness (Australia)
He might be one of those 60% guys.

He has a lot of classes and subclasses if i understand well and the low 
level functions simply get optimized by microsoft compiler far superior than 
the GCC compiler. Further details please ask him.

Also ask some details to Gian-Carlo Pascutto about the huge speedup to 64 
bits and even more with visual c++ : gcp at sjeng.org  (Belgium)

If i'm correct the free sjengr version will hit specint2006 (though it'll 
probably be specint2010 at the time it releases).
The free sjeng one is a 32 bits program however where i helped with too at 
the time.

The current commercial version is called Deep Sjeng and is not only parallel 
but far superior and profits a lot from 64 bits and visual c++ 2005.

My own experiences in user friendlyness of visual c++ 2005 aren't there yet,
as it seems to not capable for me to compile AMD64 optimized within GUI with
PGO and at command line it can easily do AMD64 but i fail to see how to use 
the pgo there!

So the 10% measurement is GCC+PGO versus "32 bits pgo'ed generic x86" 
executable produced by visual c++ 2005.

Measurements vary from ply to ply as the memory controller gets stressed 
more of course when searching deeper trees.
I've measured 10.2% to 10.6%.

That's still pretty good from GCC that it's only 10% slower at this still 
relative brandnew hardware.
GCC of course has problems adapting to newer instruction sets with more 
registers. Competing with a team of like 16 volunteers
against a couple of hundreds fulltime professionals is not so easy, so they 
still did do a good job IMHO if you look at the raw
numbers.

Those microsoft programmers seem to have been busy zero seconds to optimize 
real well for AMD so far and just focussed upon EM64T.

Vincent

----- Original Message ----- 
From: "Douglas Eadline" <deadline at clustermonkey.net>
To: "Vincent Diepeveen" <diep at xs4all.nl>
Cc: "Ed Karns" <edkarns at firewirestuff.com>; <beowulf at beowulf.org>
Sent: Monday, January 23, 2006 11:01 PM
Subject: Re: [Beowulf] Re: Q&A: BeOS and IBM Cell processor


>> Now 10% is still not much. I do know there is colleges of mine where it 
>> is
>> actually 60% speedwin under windows.
>>
> Could you provide some detail to this claim? Without data such claims are
> worthless.
>
>
> --
> Doug
>