Computer Chess Club Archives


Search

Terms

Messages

Subject: Engine testing: Memory Speed vs NPS at 400 & 32mb hash settings

Author: Aaron Gordon

Date: 08:25:46 08/08/02


The testing was done on my AthlonXP 1900+ (1.6GHz) at 1.8Ghz, 150fsb(300DDR).
I set it to 1.4gb/s and 2.2gb/s (just over a 57% memory speed increase) for the
test.

ChessTiger 14(192mb hash): 1.4gb/s 608kn/s, 2.2gb/s 671kn/s, 10.362% difference
ChessTiger 14(24mb hash): 1.4gb/s 681kn/s, 2.2gb/s 729kn/s, 7.05%

CB* Crafty 18.15 (288mb hash): 1.4gb/s 1005kn/s, 2.2gb/s 1073kn/s, 6.766%
CB* Crafty 18.15 (30mb hash): 1.4gb/s 1066kn/s, 2.2gb/s 1133kn/s, 6.285%

Deep Junior 7 (400mb hash): 1.4gb/s 1215kn/s, 2.2gb/s 1243kn/s, 2.3%
Deep Junior 7 (32mb hash): 1.4gb/s 1236kn/s, 2.2gb/s 1273kn/s, 3.0%

Fritz7 (400mb hash): 1.4gb/s 991kn/s, 2.2gb/s 1060kn/s, 6.963%
Fritz7 (32mb hash): 1.4gb/s 1043kn/s, 2.2gb/s 1094kn/s, 4.89%

Hiarcs8 (400mb hash): 1.4gb/s 195.6kn/s, 2.2gb/s 216kn/s, 10.43%
Hiarcs8 (32mb hash): 1.4gb/s 229.8kn/s, 2.2gb/s 242kn/s, 5.31%

* Chessbase version

Testing method:
I used the position "r4b1k/pp3Ppp/n1p5/2q1B3/2r5/6RP/PP2Q1P1/5R1K w - - 0 1"
and let it run until it hit a specific ply and then I wrote down the
nodes/second by just taking the node count and dividing it by the time (not
using the Chessbase kn/s readout). To slow my ram down I decreased the timings
to the lowest possible settings in the bios.

For those of you who like to visualize things.. I made two graphs for you. I'm
inexperienced at MS Excel so I don't yet know how to make it show the entire
graph (from 0 to the maximum result), it scaled it in close automatically.
Anyway... here they are.

400mb hash setting:
http://speedycpu.dyndns.org/chess/engine-memdiff57p-400h.jpg

32mb hash setting:
http://speedycpu.dyndns.org/chess/engine-memdiff57p-32h.jpg

If you guys are interested in seeing some results I'd be happy to do some
hash size testing with various engines to see how much slower/faster it is at
finding specific moves.

To give you some idea of ram speeds.. Most average SDRAM based systems with your
average 1GHz CPU get around 700-800mb/s in the test I used (Sisoft Sandra 2002
Pro ( http://www.sisoftware.demon.co.uk/sandra/ )). A very fast SDRAM based
system (KT133a, full tweaks, nearly 150fsb, etc) can pull right around 1.1gb/s.
Average DDR systems (AMD 760, KT266, some SiS chipsets) pull around 1.2-1.4gb/s.
Higher end systems (KT266a, KT333, KT400) can (with full tweaks) do 2.0gb/s
(133fsb) to 3.1gb/s (200fsb(400DDR)).

I'm not sure how much more of an increase you'd get going over the 57% I tested
but IMHO you'd see a very nice boost going from an older SDRAM box to a newer
DDR system. If you're thinking a 10% boost may not be a lot.. well, as far as
ELO goes perhaps not. Think about this though. If you took an AthlonXP 1900+ at
1.6GHz and tacked on that extra 10% in nodes/second it would be like you're
running at 1760MHz, about in the middle of an AthlonXP 2100+ & 2200+! As far as
other things go that require a lot of memory bandwidth (like Quake3) you'd see
at least a 25% framerate increase going from 1.4gb/s to 2.2gb/s as long as your
video card isn't limiting you via the fillrate. This would mean a 1.4gb/s
AthlonXP @ 2GHz (2500+) = AthlonXP 1900+ 1.6GHz @ 2.2gb/s. One thing that makes
me drool is knowing some people out there are running their AthlonXP 2200+'s at
230FSB(460DDR, 3.68gb/s max bandwidth, on KT333 you'd see 3.5gb/s real) at
2.3GHz. Ahh!! :)

Anyway.. Hope this has been somewhat useful. I love testing so if anyone
has any ideas for anything feel free to throw'em my way and I'll see what
I can do about getting that tested.



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.