Author: leonid
Date: 19:02:42 09/27/01
Go up one level in this thread
On September 26, 2001 at 14:25:50, Gian-Carlo Pascutto wrote: >I have been wondering about my engines speed lately. > >Right now, I do about 110 knps in the middlegame on >an Athlon 1000. I can say nothing about your question but it is curious for me. If you could send to me some position (as heavy as you can) and say me what NPS you have for searching the mate by brute force (if you have this kind of search) then it could be useful for responding to my old question. Also indicate your NPS for usual chess game for that position. For a long time I tried to find what is the correspondence between NPS and speed of move generator, or even program basic speed. At my opinon, only brute force search give the best comparaison. The most strange and surprising NPS correspondence I found when looking into NPS numbers for Rebel 10 and my program. Rebel is also written for DOS and on Assembler and for this reason best for looking into it. When Rebel search mate its NPS goes very often 10 times higher that mine. When Rebel look for "normal move", by brute force, its NPS is almost identical to mine. Difference is only in 30 or 40%. This time mine NPS is slightly higher. I never found any good explanation to what I saw. Leonid. >Is that good or bad? Well, everybody else (save the one >engine that shall not be named) seems to be doing more, >so it can't be very good ;) > >Programs like Crafty do about 700 knps on the same >hardware. ExChess (which has a similar design to >my engine) does about 250-350 knps. > >I couldn't care less about the actual NPS value, but >I do seem to get outsearched sometimes, which is quite >annoying. Especially since my eval isn't of the most >sophisticated either. Makes me shiver about the idea >of adding some of the more expensive terms. > >A simple explanation would be that I just suck at >programming and my program is slow. That would be >plausible, if it weren't for 2 facts: > >a) with material only eval I do +- 300 knps >b) I 'perft' a few percents faster than crafty > >I did a few more experiments (no SEE, futility pruning) >and I hardly go over about 350 knps. That's still >half the speed of crafty, with _everything_ enabled. > >I'm baffled as to where all those NPS come from/disappear >to. The only area where Crafty would hold the advantage is >capture-only movegen. > >My standard movegen does about 450 000 calls per second (testpos: LCT2 pos1). >Capture only makes that 600 000 calls per second. I could >see how crafty is faster here. (anyone care to test how >much?) However, I don't see how this could be enough to >compensate for the huge difference even without an eval >on my part! Crafty is _searching_ faster than I can generate >moves! > >So, basically, where are all those NPS coming from? > >You guys have a 'nodes <<= 2;' somewhere? > >-- >GCP
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.