Computer for chess analysis

Main

Date: 06 Feb 2006 00:41:38
From: Ange1o DePa1ma
Subject: Computer for chess analysis

I'm aware that the major programs all suffer from a horizon effect, which as
I understand it means they cannot see beyond a certain move. I demonstrated
this is so on my machine with a certain endgame position that required a
queen exchange and outflanking over 7-8 moves. I saw this sortie and played
it, but Shredder 7 takes quite a bit of time (on my machine) to see the
winning move.

Questions:

1. Is there any way to increase an engine's horizon or calculating speed
(this may be two questions)?

2. If you could design a dedicated chess computer running under Windows,
what would the components be and what would the settings be? This computer
would have no other programs running other than system stuff and would not
be connected to the Internet.

Thanks,

Angelo

Date: 07 Feb 2006 08:39:27
From: Johnny T
Subject: Re: Computer for chess analysis

Ange1o DePa1ma wrote:

> 1. Is there any way to increase an engine's horizon or calculating speed
> (this may be two questions)?
No not really.

> 2. If you could design a dedicated chess computer running under Windows,
> what would the components be and what would the settings be? This computer
> would have no other programs running other than system stuff and would not
> be connected to the Internet.

Wouldn't really matter. You would get percentage point differences.
But even if you could get say a doubling, it wouldn't really make a
difference. Raw look ahead speed is a logarithmic problem. You need
magnitudes higher speed. Currently this is only truly available with
custom hardware. Like the kind that doesn't exist anymore with deep
blue, or that does exist with hydra which is a multi-million dollar
solution.

And it really looks like we are near the end of the line for desktop
speed improvements. But all is not lost, I suspect we are going to see
a reemergence of "smart" chess programs which will move forward the
strength of programs beyond what brute strength will provide. Rybka,
and Fritz9, and Hiarcs 10 are moving us there.

Date:
From: Martin Brown
Subject: Re: Computer for chess analysis

Date: 06 Feb 2006 12:39:39
From: David Vancina
Subject: Re: Computer for chess analysis

A fast processor and *lots* of fast memory are important. More memory =
bigger hash tables. Faster processor means more positions analyzed per
second. If it's not internet-connected you can strip off all the
typical antivirus and firewall stuff. Beyond these basics I've read
some interesting posts here regarding larger CPU pipelines, suggesting
that feature benefits the particular type of processing done by chess
programs, but I'm not sure if 1) it's a correct assertion; or 2) which
CPU's have this characteristic.

I suspect a lot of fiddling will get into diminishing returns territory
pretty quickly -- not much effect for the time and money invested. I'm
no expert, but I'd just go with a modern, fast CPU with the fastest FSB
clock rate they make and a gig (or 2?) of memory. The clock rate of the
Front Side Bus is, as I understand it, a determinant of the machine's
speed in accessing memory. Since chess programs are so memory
intensive, memory speed counts for a lot. You then want to make sure
you put in memory that can keep up.

As I said, I'm not an expert but this is my understanding of the issues.
If I'm wrong I'm sure someone will be good enough to point out
improvements. ;-)

DJV

Ange1o DePa1ma wrote:
> I'm aware that the major programs all suffer from a horizon effect, which as
> I understand it means they cannot see beyond a certain move. I demonstrated
> this is so on my machine with a certain endgame position that required a
> queen exchange and outflanking over 7-8 moves. I saw this sortie and played
> it, but Shredder 7 takes quite a bit of time (on my machine) to see the
> winning move.
>
> Questions:
>
> 1. Is there any way to increase an engine's horizon or calculating speed
> (this may be two questions)?
>
> 2. If you could design a dedicated chess computer running under Windows,
> what would the components be and what would the settings be? This computer
> would have no other programs running other than system stuff and would not
> be connected to the Internet.
>
> Thanks,
>
> Angelo
>
>

Date: 05 Apr 2006 18:04:23
From: Thomas T. Veldhouse
Subject: Re: Computer for chess analysis

David Vancina <djvchess@c-o-m-c-a-s-t.net > wrote:
> A fast processor and *lots* of fast memory are important. More memory =
> bigger hash tables. Faster processor means more positions analyzed per
> second. If it's not internet-connected you can strip off all the
> typical antivirus and firewall stuff. Beyond these basics I've read
> some interesting posts here regarding larger CPU pipelines, suggesting
> that feature benefits the particular type of processing done by chess
> programs, but I'm not sure if 1) it's a correct assertion; or 2) which
> CPU's have this characteristic.
>

64-bit processors with 64-bit softare (including the OS) running on them will
yield substantially improved results due to the fact that most, if not all
current chess software uses bitboard theory and bitboards are represented by a
64-bit number ... ideally suited to 64-bit processors.

Other than that. You want the processor to perform well with integer math.
You want fast memory and a fast bus to match. Last, but not least, you want a
lot of memory and then increase the hash table size accordingly in your chess
software. A side note, disk I/O will probably help, but generally, relavent
data will be cached, so it is a secondary requirement.

--
Thomas T. Veldhouse
Key Fingerprint: 2DB9 813F F510 82C2 E1AE 34D0 D69D 1EDC D5EC AED1

Date: 06 Apr 2006 10:26:15
From: David Richerby
Subject: Re: Computer for chess analysis

Thomas T. Veldhouse <veldy71@yahoo.com > wrote:
> 64-bit processors with 64-bit softare (including the OS) running on
> them will yield substantially improved results due to the fact that
> most, if not all current chess software uses bitboard theory

Is that actually true? I was under the impression that bitboards were
rather out of fashion, these days. (I have no citation for this
statement so it should be classed as `unsubstantiated rumour at best
unless anyone can confirm it.)

> A side note, disk I/O will probably help, but generally, relavent
> data will be cached, so it is a secondary requirement.

I'd have thought disc I/O speed would be fairly insignificant. It's
only going to affect tablebase lookup times and, since tablebases are
so much faster than searching anyway, the difference between a `slow'
tablebase lookup and a lightning fast one sounds insignificant. This
is certainly only an issue while the position on the board is not yet
in tablebases so the TB is only getting probed at the end of each
line. If the board position is in TB, you only need a handful of
lookups per move and it really does't matter how many tens of
milliseconds that takes.

Dave.

--
David Richerby Simple Monk (TM): it's like a man of
www.chiark.greenend.org.uk/~davidr/ God but it has no moving parts!

Date: 06 Apr 2006 12:22:54
From: Thomas T. Veldhouse
Subject: Re: Computer for chess analysis

David Richerby <davidr@chiark.greenend.org.uk > wrote:
> Thomas T. Veldhouse <veldy71@yahoo.com> wrote:
>> 64-bit processors with 64-bit softare (including the OS) running on
>> them will yield substantially improved results due to the fact that
>> most, if not all current chess software uses bitboard theory
>
> Is that actually true? I was under the impression that bitboards were
> rather out of fashion, these days. (I have no citation for this
> statement so it should be classed as `unsubstantiated rumour at best
> unless anyone can confirm it.)
>

I *could* be wrong, but I believe they are alive and well. I think they are
just augmented by specific intelligence algorithms these days.
>
>> A side note, disk I/O will probably help, but generally, relavent
>> data will be cached, so it is a secondary requirement.
>
> I'd have thought disc I/O speed would be fairly insignificant.

That is what I said. Since I/O might be a logical question to ask, I
addressed it up front, and declared it a secondary consideration. There is
I/O, but it is related to database searches.

> It's
> only going to affect tablebase lookup times and, since tablebases are
> so much faster than searching anyway, the difference between a `slow'
> tablebase lookup and a lightning fast one sounds insignificant.

Table bases are also cached typically. I believe chessbase products default
to a 1MB cache.

> This
> is certainly only an issue while the position on the board is not yet
> in tablebases so the TB is only getting probed at the end of each
> line. If the board position is in TB, you only need a handful of
> lookups per move and it really does't matter how many tens of
> milliseconds that takes.
>

I think the value of caching comes in when you approach the threshold of
tablebases (i.e. if you have only 4-man and lower table bases and the engine
is finding lines that are say 7 moves from the end of the game, the engine may
well hit tablebase calculations very often, hence, the reason they are
cached).

--
Thomas T. Veldhouse
Key Fingerprint: 2DB9 813F F510 82C2 E1AE 34D0 D69D 1EDC D5EC AED1

Date: 06 Apr 2006 14:47:17
From: David Richerby
Subject: Re: Computer for chess analysis

Thomas T. Veldhouse <veldy71@yahoo.com > wrote:
> David Richerby <davidr@chiark.greenend.org.uk> wrote:
>> Thomas T. Veldhouse <veldy71@yahoo.com> wrote:
>>> 64-bit processors with 64-bit softare (including the OS) running on
>>> them will yield substantially improved results due to the fact that
>>> most, if not all current chess software uses bitboard theory
>>
>> Is that actually true? I was under the impression that bitboards were
>> rather out of fashion, these days.
>
> I *could* be wrong, but I believe they are alive and well.

OK. Let's see if anyone can be more definitive but, for the time
being, I'll accept your belief as having higher authority than my
vague impression.

>>> A side note, disk I/O will probably help, but generally, relavent
>>> data will be cached, so it is a secondary requirement.
>>
>> I'd have thought disc I/O speed would be fairly insignificant.
>
> That is what I said.

To me, `will probably help' and `is a secondary requirement' sounds
much more than `fairly insignificant', which is why I queried what you
said. But it's clear that we agree on this, now.

>> This is certainly only an issue while the position on the board is
>> not yet in tablebases so the TB is only getting probed at the end
>> of each line.
>
> I think the value of caching comes in when you approach the
> threshold of tablebases (i.e. if you have only 4-man and lower table
> bases and the engine is finding lines that are say 7 moves from the
> end of the game, the engine may well hit tablebase calculations very
> often, hence, the reason they are cached).

Yes, that's what I was trying to say. :-)

Dave.

--
David Richerby Artificial Pants (TM): it's like a
www.chiark.greenend.org.uk/~davidr/ well-tailored pair of trousers that's
made of plastic!

Date: 07 Feb 2006 01:42:44
From: Ange1o DePa1ma
Subject: Re: Computer for chess analysis

Thanks!

"David Vancina" <djvchess@c-o-m-c-a-s-t.net > wrote in message
news:Ao-dnQiVuddrCnreRVn-gw@comcast.com...
>A fast processor and *lots* of fast memory are important. More memory =
>bigger hash tables. Faster processor means more positions analyzed per
>second. If it's not internet-connected you can strip off all the typical
>antivirus and firewall stuff. Beyond these basics I've read some
>interesting posts here regarding larger CPU pipelines, suggesting that
>feature benefits the particular type of processing done by chess programs,
>but I'm not sure if 1) it's a correct assertion; or 2) which CPU's have
>this characteristic.
>
> I suspect a lot of fiddling will get into diminishing returns territory
> pretty quickly -- not much effect for the time and money invested. I'm no
> expert, but I'd just go with a modern, fast CPU with the fastest FSB clock
> rate they make and a gig (or 2?) of memory. The clock rate of the Front
> Side Bus is, as I understand it, a determinant of the machine's speed in
> accessing memory. Since chess programs are so memory intensive, memory
> speed counts for a lot. You then want to make sure you put in memory that
> can keep up.
>
> As I said, I'm not an expert but this is my understanding of the issues.
> If I'm wrong I'm sure someone will be good enough to point out
> improvements. ;-)
>
> DJV
>
> Ange1o DePa1ma wrote:
>> I'm aware that the major programs all suffer from a horizon effect, which
>> as I understand it means they cannot see beyond a certain move. I
>> demonstrated this is so on my machine with a certain endgame position
>> that required a queen exchange and outflanking over 7-8 moves. I saw this
>> sortie and played it, but Shredder 7 takes quite a bit of time (on my
>> machine) to see the winning move.
>>
>> Questions:
>>
>> 1. Is there any way to increase an engine's horizon or calculating speed
>> (this may be two questions)?
>>
>> 2. If you could design a dedicated chess computer running under Windows,
>> what would the components be and what would the settings be? This
>> computer would have no other programs running other than system stuff and
>> would not be connected to the Internet.
>>
>> Thanks,
>>
>> Angelo