Reply
 
LinkBack Thread Tools Display Modes
  #1   Report Post  
Old March 21st 04, 06:00 PM
Jan
 
Posts: n/a
Default Computer speed

Hi all

I'm trying to determine what computer configurations have a good chess
performance / cost ratio.
I installed Fritz 8 everywhere, and let Fritz analyse the position after
1.e4, 16 plies deep, one line
I have following results:

Xeon 2.4 GHz (5) 1024 MB XPPro 128 MB 806 16/44 0:03:34 173213
Xeon 2.4 GHz (4) 1024 MB XPPro 256 MB 802 16/48 0:04:23 211149
Xeon 2.4 GHz (3) 1024 MB XPPro 512 MB 795 16/47 0:03:35 171033
Xeon 2.4 GHz (6) 1024 MB XPPro 818 MB 793 16/46 0:03:53 185134
AMD Athlon XP 2200+ (1) 256 MB NT2000 sp3 192 MB 722 16/47 0:04:02 174324
Intel P4 1.8 GHz (2) 256 MB 2000 96 MB 586 16/44 0:05:54 207941
AMD XP-M 2200+ 192 MB XPHome 96 MB 486
AMD Duron 1 GHz 256 MB 2000 P 96 MB 456

Sorry about the formatting. http://users.skynet.be/fa096445/tw/Fritz8.txt
should look better

1. Is this a good test?
2. Why is my XP-M laptop so slow?
3. Why does the engine become slightly slower by increasing hash size?
4. Does anyone want to share his results too?

Kind regards
Jan


  #2   Report Post  
Old March 22nd 04, 01:42 PM
Euclid
 
Posts: n/a
Default Computer speed

This might be interesting but I have no idea what your numbers mean. Please
explain and define. Thanks. I'll just mention that with a test like this,
you should use the same hash size with each machine.

Another easy test... Run infinite analysis mode on the starting chessboard
position, and just make a note of the kN/s (speed) after a fixed amount of
time (say 3 minutes, for example). That is a quicky comparison, and pretty
good IMHO. Again for proper comparison the hash size should be the same on
each machine.

At least that's what I did when comparing some different machines last
summer before deciding on a model to purchase. An AMD Athlon XP2400+
computer won by a large margin (2x faster, or more) in my comparisions
versus Intel Celerons in the same low-price category (computers costing
under $500). So I bought it from Walmart (S4020WM) then put in more RAM and
a bigger hard drive, and am happy with it. A significant positive factor was
the computer case, big, easy to open, well-laid out, easy to access and work
inside.
-E

"Jan" wrote in message
...
Hi all

I'm trying to determine what computer configurations have a good chess
performance / cost ratio.
I installed Fritz 8 everywhere, and let Fritz analyse the position after
1.e4, 16 plies deep, one line
I have following results:

Xeon 2.4 GHz (5) 1024 MB XPPro 128 MB 806 16/44 0:03:34 173213
Xeon 2.4 GHz (4) 1024 MB XPPro 256 MB 802 16/48 0:04:23 211149
Xeon 2.4 GHz (3) 1024 MB XPPro 512 MB 795 16/47 0:03:35 171033
Xeon 2.4 GHz (6) 1024 MB XPPro 818 MB 793 16/46 0:03:53 185134
AMD Athlon XP 2200+ (1) 256 MB NT2000 sp3 192 MB 722 16/47 0:04:02 174324
Intel P4 1.8 GHz (2) 256 MB 2000 96 MB 586 16/44 0:05:54 207941
AMD XP-M 2200+ 192 MB XPHome 96 MB 486
AMD Duron 1 GHz 256 MB 2000 P 96 MB 456

Sorry about the formatting. http://users.skynet.be/fa096445/tw/Fritz8.txt
should look better

1. Is this a good test?
2. Why is my XP-M laptop so slow?
3. Why does the engine become slightly slower by increasing hash size?
4. Does anyone want to share his results too?

Kind regards
Jan



  #3   Report Post  
Old March 23rd 04, 02:13 AM
Simon Waters
 
Posts: n/a
Default Computer speed

Jan wrote:

1. Is this a good test?


What Euclid said

2. Why is my XP-M laptop so slow?


Pass - what's the detailed processor spec?

A quick glance suggests the processor is optimised for low power
consumptions and running Windows XP and Office apps, not playing chess.

3. Why does the engine become slightly slower by increasing hash size?


Hash quite often slows chess computers down.

The performance killer in most chess programs is memory access.

The more memory it has the more time it spends writing stuff to and
(especially) fetching from memory, instead of thinking.

Of course if you remember everything, you NEED think less as you'll say
"ah I've thought about this before". But even remembering if you thought
of something before, requires checking in memory, which is slow.

A good example is GNU Chess on WAC (win at chess test positions), if I
switch hashing off it solves a different selection of puzzles at 5
seconds a move. With hashing it can solve complex endgames with myriad
transpositions (even quite a small hash makes a huge difference),
without it it solves complex tactical positions better.

Against skilled human opposition skill at end games counts far more than
tactical prowess (where a cheap computer from the early 1980's can beat
the best humans), so often it is a good trade off - memory for speed.

Between processors with similar architectures the size of the CPU cache
is something to look out for. Intel Xeon offers a range of caches upto
4.5MB (L2+L3), which is about 7 times the cache of the AMD XP-M, and
offers twice the bus speed (and presumably faster memory access). But
that is the difference between a big server processor and a laptop
processor, I dare say AMD will sell you a big server processor to
compete with Intel's.

In the archive here you'll find some interesting comments from Robert
Hyatt, which discuss how processors fetch stuff from memory, how long it
takes, how big a chunk they fetch at a time, and what this means when
designing chess programs. If it all sounds complex - try thinking about
what happens when you have two processors sharing the same memory -
boggles my mind.

Try working out how big the CPU cache is, and see if setting the hash
size around that size makes a dramamtic difference to speed and/or skill
on a set of test positions.

-----BEGIN PGP SIGNATURE-----
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFAX502GFXfHI9FVgYRApbAAJ9qwOUeUoSWyyzjt/Lw24CtW13RHgCeM76V
1Yd+rSZKtR18qLEcCaKedAI=
=lTJn
-----END PGP SIGNATURE-----

  #4   Report Post  
Old March 23rd 04, 12:45 PM
David Richerby
 
Posts: n/a
Default Computer speed

Simon Waters wrote:
Try working out how big the CPU cache is, and see if setting the hash
size around that size makes a dramamtic difference to speed and/or skill
on a set of test positions.


The OP's data suggest that the hash table sweet spot is somewhere around
256Mb -- that's orders of magnitude bigger than the CPU cache.


Dave.

--
David Richerby Hungry Chocolate Hi-Fi (TM): it's
www.chiark.greenend.org.uk/~davidr/ like a music system that's made of
chocolate but it'll eat you!
  #5   Report Post  
Old March 23rd 04, 10:14 PM
Simon Waters
 
Posts: n/a
Default Computer speed

David Richerby wrote:
Simon Waters wrote:

Try working out how big the CPU cache is, and see if setting the hash
size around that size makes a dramamtic difference to speed and/or skill
on a set of test positions.



The OP's data suggest that the hash table sweet spot is somewhere around
256Mb -- that's orders of magnitude bigger than the CPU cache.


All I'm suggesting is the behaviour at several orders of magnitude may
also be interesting, and may better reveal other aspects of the hardware.

-----BEGIN PGP SIGNATURE-----
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFAYLauGFXfHI9FVgYRArOaAJ4tZETEOU22dHUrKFXoPQ j8Hsj6xACcDIXB
XcUSoAyCaCBeepWwTY7JduU=
=+HAN
-----END PGP SIGNATURE-----



  #6   Report Post  
Old March 26th 04, 02:08 AM
David
 
Posts: n/a
Default Computer speed

On Sun, 21 Mar 2004 19:00:01 +0100, "Jan" wrote:

Hi all

I'm trying to determine what computer configurations have a good chess
performance / cost ratio.
I installed Fritz 8 everywhere, and let Fritz analyse the position after
1.e4, 16 plies deep, one line
I have following results:

Xeon 2.4 GHz (5) 1024 MB XPPro 128 MB 806 16/44 0:03:34 173213
Xeon 2.4 GHz (4) 1024 MB XPPro 256 MB 802 16/48 0:04:23 211149
Xeon 2.4 GHz (3) 1024 MB XPPro 512 MB 795 16/47 0:03:35 171033
Xeon 2.4 GHz (6) 1024 MB XPPro 818 MB 793 16/46 0:03:53 185134
AMD Athlon XP 2200+ (1) 256 MB NT2000 sp3 192 MB 722 16/47 0:04:02 174324
Intel P4 1.8 GHz (2) 256 MB 2000 96 MB 586 16/44 0:05:54 207941
AMD XP-M 2200+ 192 MB XPHome 96 MB 486
AMD Duron 1 GHz 256 MB 2000 P 96 MB 456

Sorry about the formatting. http://users.skynet.be/fa096445/tw/Fritz8.txt
should look better

1. Is this a good test?
2. Why is my XP-M laptop so slow?
3. Why does the engine become slightly slower by increasing hash size?
4. Does anyone want to share his results too?

Kind regards
Jan


I don't believe your chess position is very representative, since only
one pawn has been moved.

Fritz has a means to measure CPU/system speed, called Fritzmark's.
That would be a better standard to follow.

The laptop probably has a feature like my laptop, which may slow the
processor down as it believes it should, to conserve power.

Was the laptop running on batteries or from a wall socket power
source, when tested?

dave
Meter by meter, life is sweeter - go metric, save millions, and take it easy.
  #7   Report Post  
Old March 27th 04, 08:31 AM
Jan
 
Posts: n/a
Default Computer speed


"David" schreef in bericht
...
On Sun, 21 Mar 2004 19:00:01 +0100, "Jan" wrote:

Hi all

I'm trying to determine what computer configurations have a good chess
performance / cost ratio.
I installed Fritz 8 everywhere, and let Fritz analyse the position after
1.e4, 16 plies deep, one line
I have following results:

Xeon 2.4 GHz (5) 1024 MB XPPro 128 MB 806 16/44 0:03:34 173213
Xeon 2.4 GHz (4) 1024 MB XPPro 256 MB 802 16/48 0:04:23 211149
Xeon 2.4 GHz (3) 1024 MB XPPro 512 MB 795 16/47 0:03:35 171033
Xeon 2.4 GHz (6) 1024 MB XPPro 818 MB 793 16/46 0:03:53 185134
AMD Athlon XP 2200+ (1) 256 MB NT2000 sp3 192 MB 722 16/47 0:04:02

174324
Intel P4 1.8 GHz (2) 256 MB 2000 96 MB 586 16/44 0:05:54 207941
AMD XP-M 2200+ 192 MB XPHome 96 MB 486
AMD Duron 1 GHz 256 MB 2000 P 96 MB 456


Sorry for my late responses.

First column: processor type. Followed by installed memory, OS version,
hash size, kN/s, search depth, search time, knodes searched.

I don't believe your chess position is very representative, since only
one pawn has been moved.

Fritz has a means to measure CPU/system speed, called Fritzmark's.
That would be a better standard to follow.


I noticed some days ago. It is indeed a far better way to do the
benchmarking. What I was trying to do was to try to determine good
processors and configurations. Trying to be able to estimate how much
faster a PC would be by increasing memory size, ...


The laptop probably has a feature like my laptop, which may slow the
processor down as it believes it should, to conserve power.


Actually, it is increasing speed at full load. Idle, it runs at +- 800 MHz,
and stressed at +-1100. The search speed seems to be proportional to
1100/1000, compared to my Duron 1 Ghz.


Was the laptop running on batteries or from a wall socket power
source, when tested?


Interesting question, it was running from the power supply. I should test
the difference.

I'll try to find a site displaying some recent Fritzmark tests.

Kind regards and thx for all responses
Jan


Reply
Thread Tools
Display Modes

Posting Rules

Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On


Similar Threads
Thread Thread Starter Forum Replies Last Post
(C)heating Revisted Gunny Bunny rec.games.chess.computer (Computer Chess) 12 March 13th 04 08:20 PM
Computer vs. Computer Servers Neko rec.games.chess.computer (Computer Chess) 2 November 26th 03 09:14 PM
chess computer website Chris Kantack rec.games.chess.computer (Computer Chess) 1 November 12th 03 09:42 PM
1980's chess computer with MOVING pieces. SmartBloke rec.games.chess.computer (Computer Chess) 2 September 18th 03 03:51 PM
learning chess: how to use the computer opponent? Bob rec.games.chess.computer (Computer Chess) 12 September 15th 03 09:27 AM


All times are GMT +1. The time now is 11:39 AM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.
Copyright 2004-2019 ChessBanter.
The comments are property of their posters.
 

About Us

"It's about Chess"

 

Copyright © 2017