Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Subject: Re: Some Chess benches .. ( Crafty data)

Author: Drexel,Michael

Date: 15:31:40 08/07/04

Go up one level in this thread


On August 07, 2004 at 16:11:28, Peter Berger wrote:

>First question was if the Crafty bench produced meaningful data.
>
>I decided to test this with crafty.rc's I'd use for a match. With the PIV3.2,
>that has 1GB RAM this meant
>
>smpmt=1
>hash 384M
>hashp 48M
>cache 32M
>noise 500000
>tbpath=C:\tbs
>swindle off
>computer
>egtb
>resign 6
>exit
>
>On the Athlon 64 3200+ I used half of this value for hash and hashp as it has
>only 512 MB RAM.
>
>To also answer Gerd's theory ( although HT was disabled in previous test) I did
>another run with smpt=2 on the PIV.
>
>And to Bob: a 64 bit executable would be of limitted use for me - also if you
>first had to lose 20% only to recover it with the 64bit compile .. :)
>
>Test was done with the crafty 19.15 SE executable provided by Mike Byrne.
>
>quick.epd (60s)
>
>        Athlon 64                PIV3.2(mt=1)            PIV3.2(mt=2)
>
>#1      11/-   932314           11/-   1094865          11/-
>#2      13/-  1231821           13/-   1446202          13/-
>#3      16/+  1305247           17/+   1587329          15/+
>#4      13/-   822912           13/-   1084778          13/-
>#5      13/+  1004094           12/+   1221268          12/+
>#6      14/+   961920           14/+   1127484          14/+
>#7      13/+  1195035           13/+   1374959          13/+
>#8      15/-  1069003           15/-   1244538          15/-
>#9      12/+   806188           13/+   1045365          11/+
>#10     12/-   937916           13/-   1182559          13/-
>#11     11/-   962095           11/+   1222099          11/+
>#12     12/-   749055           12/-    964936          12/-
>#13     12/+  1164411           12/+   1422361          12/+
>#14     13/+  1005028           14/+   1209125          14/+
>#15     13/+  1158575           13/+   1380033          13/+
>#16     8/+    629500            7/+    719015          11/+
>#17     12/+   875415           12/+   1134823          12/+
>#18     13/+   957280           13/+   1202340          13/+
>#19     14/+  1174810           14/+   1348973          14/+
>#20     11/+   744611           11/+    960048          11/+
>#21     10/+   744877           11/+    949471          11/+
>#22     14/+  1151032           15/+   1348936          14/+
>#23     12/+  1094375           12/+   1341502          12/+
>#24     15/+  1090312           15/+   1296615          15/+
>
>Avg+Sol 12.58/17 990326         12.75/18 1204568        12.71/18
>
>
>The format here is depth reached, solved, Nodes/s. The result shows that the PIV
>is indeed about 20% faster with this executable.
>
>HT doesn't provide a benefit.
>
>The results with the ECMGCP testsuite confirm both findings:
>
>
>ECMGCP (10s)
>
>1 s      56/183                   59/183                57/183
>2 s      75                       83                    77
>5 s      101                      107                   106
>10s      123                      126                   126
>
>So the benches produce reasonable results, and Bryan Hoffman's explanation looks
>like a good guess: maybe Mike's executable just sucks on the Athlon64 ?
>
>Previous bench results with the SE executable (hash 48M, hashp 12M, nothing
>else):
>
>Athlon 64 : "Mike's" Crafty 19.15  Total nodes: 58028353             1115929
> 52 12.307692
>PIV 3.2   : "Mike's" Crafty 19.15  Total nodes: 58028353 Raw nodes/s:1349496
>Time:43 14.884
>
>Now I added Dann Corbit's executable to the test:
>
>Athlon 64: "Dann's" Crafty 19.15  Total nodes: 63988715  Raw nodes/s:1361462
>Time:47 13.617

I guess you have an Athlon 64 3200+ 2.0 Ghz, NOT 2.2Ghz.
That would explain everything.

Here are the results (19.15 Dann,48M hash) for Athlon XP 2.4Ghz and 2.5Ghz:

Total nodes: 63988715
Raw nodes per second: 1421971
Total elapsed time: 45
SMP time-to-ply measurement: 14.222222

Total nodes: 63988715
Raw nodes per second: 1488109
Total elapsed time: 43
SMP time-to-ply measurement: 14.883721

An Athlon 64 3200+ 2.2 Ghz should get about 1.5M N/s with this 32-bit
executable.

Michael


>
>
>I actually did some more tests, but I am too tired, so just the conclusions.
>Basically Mike's executable is just great for the PIV but bad for the Athlon.
>The PIV is a little faster anyway but it is in the region below 10%.
>
>Peter



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.