Author: Dan Newman
Date: 01:31:52 07/23/00
Go up one level in this thread
On July 22, 2000 at 20:52:39, Larry Griffiths wrote:
>On July 22, 2000 at 16:07:57, Dan Newman wrote:
>>
>>Mine gets about 1.355 million nodes/s on a 400 MHz Celeron on this test.
>>To compare with a 550 Xeon multiplying by 550/400 might be close enough:
>>1.3 x 550/400 == 1.86 M/s, which is pretty close to your number.
>>
>>It could be a matter of terminology. Some people were (in some threads)
>>comparing move generation rate in which you call the move generator
>>repetitively without any search. This is generally what I'm talking about
>>if I say "moves per second". Or it could be a matter of faster processors.
>>
>>The above (brute force) test is weighted towards measuring the speed of
>>the make/undo code since you end up calling make/undo about 30x for each
>>move generation--but make/undo speed might be slightly more important
>>than move gen speed anyway... (This test also equally mixes non-capture
>>with capture generation, but the capture generation speed is much more
>>important usually.)
>>
>>Anyway, it looks like you're going fast enough :)
>>
>>-Dan.
>
>Thanks for the Info Dan.
>My 550 machine is 1.5 times faster than the 550 that I use at work.
>The Pentium III Xeon has a 512k Level 1 cache that runs at the same speed as the
>cpu. This makes it more like a 825 Mhz machine so using your formula...
>
>1.355 x 825/400 == 2.79 M/s
>
>I did a little optimization and have it up to 2.1 million moves per second at
>this time.
>
>I do not have code to validate the moves yet, and I also do not have
>code for en-passant, castleing, or pawn promotion.
>
>I assume that your 1.355 figure includes code for en-passant, castleing, and
>pawn promotion? Am I correct?
>
Yes, it's got all that and is validating moves and doing in-check detection,
etc., as well.
I thought the Celeron was more or less a Xeon but with a much smaller (128k)
cache. My code tends to run a lot faster on the Celerons than on PIII's
(with the half speed 512k cache). I went to a store recently and benchmarked
a few machines with my program and got the following:
Cel/400: 424 knps 1.00 x Cel/400
P3/500: 472 knps 1.11
Cel/533: 555 knps 1.31
P3/600 (512kB): 555 knps 1.31
Cel/667: 672 knps 1.58
Athlon/800 698 knps 1.65
It apparently likes Celerons and doesn't like old PIII's or Athlons. (The
Celerons have the slower 66 MHz buses too.)
I don't know how it does on the newer 256k PIII's though...
-Dan.
>Larry :)
This page took 0 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.