Author: Robin Smith
Date: 22:40:19 01/28/99
Go up one level in this thread
On January 28, 1999 at 15:40:13, Robert Hyatt wrote: >On January 27, 1999 at 18:07:40, Robin Smith wrote: > >>Has anyone compared the "nodes per second" search rate for the various chess >>programs out there? The reason I am curious is it seems to me that for equaly >>rated programs, "fast searchers" might excell at speed chess and/or tactical >>positions, while perhaps "slow searchers" will perform better at very long time >>controls and/or more strategic positions. Has anyone made this comparison and >>are my assumtions reasonable? > >No... because it is the wrong way to compare programs. It is great to >compare the same program to a modified version of itself, if your intent was >to optimize a bit of the code... because with identical programs, faster is >better. > >But take two automobiles. My son has a 1990 ford Mustang, with a 3.08:1 >final drive ratio. Hot little car. In 1968, I had a plymouth road runner >with a 4.10:1 final drive ratio. If we could race them, and only use the >tachomoter to pick the winner, my roadrunner would win easily. If we look >at the speedometer, the mustang would win, because my roadrunner would run >out of gears too quick, and probably wouldn't do much over 120-130mph, while >that mustang will likely get real close to 150. > >But if we had two mustangs (identical) then watching the tach would work, >because the one turning the most rpm would be going the fastest, since >everything (including final drive ratio) is equal. > >That's the pitfall of looking only at NPS when looking at different programs. >His motor might be turning slower, but he might be going faster for all we >know if he is very selective... I have read all the reponses to my orriginal question and it looks like NPS is not a good way to compare programs. Let me ask directly what I really want to know. The SSDF ratings list is great for telling me how the various programs compare when playing against other computers under certain time controls. But this doesn't tell all that much about the strengths and weaknesses of the various chess engines. If I have several programs, how can I decide which one will be most likely to give good results under different conditions than the SSDF ratings list? For example, maybe I want a very fast blunder checking program to search through large databases and only spening at most a couple seconds per move. Which program(s) would likely get the best overall result when doing fast blunder checking? Or say I am going over some GM game and don't understand why a certain move was played. Which program might be best at thinking about the position over night and rendering a verdict? Which programs excell at tactics? Which tend to understand quiet, positional games better? etc. I would really like a more detailed description of the relative strengths and weaknesses of chess engines under different circumstances. SSDF doesn't do that. Does anything? Robin Smith
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.