Author: Kai Skibbe
Date: 05:18:28 12/14/98
Go up one level in this thread
On December 14, 1998 at 08:11:29, Didzis Cirulis wrote: >>Would you specify this? > >Here I repost my message posted december 4 as reply to Kai Skibbe: > >So, What do we have - a single PC and any two programs. And we are mad with >desire to test the programs we have got as fair as possible. Right? > >Step one: make sure both programs can be run with the permanent brain off. For >example, it is difficult for Nimzo 98 as there is no PB on/off option in this >program. > >Step two: For dos programs make sure they get as much processor power as >possible when active and gets suspended when unactive. > >Step three: find out the Hash sizes your programs can afford. Example: On my 32 >Mb RAM system, I usually give 8 Mb of hash to both programs to avoid any >swapping. This IS important, as swaping makes everything useless. > >Step four: Imagine there is an amount of calculations that should be done by a >program, say, 1 milion positions. And now you should do some testing: > Situation (all numbers are given as an example only!): > > 1) get the time used by the program A when it runs alone: 100 seconds > 2) same task, same program A, but program B is there > unactive in the background, and hogs some computing power: 140 seconds > > 3) get the time used by the program B when it runs alone: 120 seconds > 4) same task, same program B, but program A is there > unactive in the background, and hogs some computing power: 125 seconds > > So, in both situations it may happen that program runs slower if another one >sits there in the background. > >Step five: Calculations! > Formula: > koeficient =(result with a "hog" in background)/ (result if alone) > > For program A it makes: 140/100=1.4 > For program B it makes: 125/120=1.042 > >Step six: Let's say we want to play at 30 minutes per game level. What do we do >now is the calculation of the times needed by both programs: > > For program A: 30 minutes x 1.4 = 42 minutes > For program B: 30 minutes x 1.042 = 31.26 minutes or let's say 31 minute. > >Final step: we give Program A the 42 minutes for the game, and 31 minute to >Program B, and :-) > >Run all tests 3 times! Sometimes it is required to restart the program every >time to be sure we get the same situation again. > >This is the way we can make a fair match even on a single PC where both programs >are busy stealing the computing power from each other :-) > >There may be different practical applications of this. One program may have some >built-in test that is very convenient for this purpose, but another one may be >tested running a test position at a fixed depth. Or there may be programs where >you should disable Book, and run the initial position as a test position... Ask >me if any problems. > >Didzis > Hi Didzis, for Rebel10 and CM6000 this worked just fine. I have calculated a factor of about 1.3 for Rebel10. So I play these two programs with 60min/game for CM6000 and 75min/game for Rebel10. Junior has a factor of about 3.6 and Fritz5.32 a factor of about 8. So I decided to play with Rebel10 first to get a first impression of the strength of CM6000. Best regards Kai
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.