Author: Robert Hyatt
Date: 19:05:58 07/23/00
Go up one level in this thread
On July 23, 2000 at 18:08:17, Eelco de Groot wrote: >On July 23, 2000 at 01:52:03, Jeremiah Penery wrote: > >>On July 22, 2000 at 16:26:19, Robert Hyatt wrote: >> >>>On July 22, 2000 at 14:44:19, Jeremiah Penery wrote: >>> >>>>On July 22, 2000 at 09:35:12, Robert Hyatt wrote: >>>> >>>>>This is way too hard to answer directly, because it varies by program and >>>>>by approach. My NPS doesn't change much, because of the way my move generator >>>>>works. I have noticed that Ferret is just the opposite and speeds up >>>>>significantly in the endgame. >>>> >>>>On my machine (PII-300), the NPS of Crafty goes from about 90k in the opening to >>>>almost 300k sometimes in the ending. >>> >>> >>>I don't see swings that big. With tablebases my NPS gets pretty low in >>>endings... >> >>True, as I'm not using any TBs at the moment. When I had TBs enabled, my NPS >>dropped a ton, partly because I have a pretty slow HD. IIRC, the NPS dropped to >><50k often when about 8-9 pieces were still on the board. > >I always thought that the main speed-up came about because, when there are only >a few pieces left on the board, the number of transpositions goes up, especially >for non-pawns. At least that is what I understood from a non-scientific article >in Computerschaak. Leonid should be able to check that with his program if he >has a way to check the number of hashtable-hits. So I thought that in that case >then the NPS does not say very much because you would be looking at the same >position reached in different ways (Qa1-a6-b6, Qa1-b1-b6 etc.) Most of these >move-orders probably amount to the same thing.. And if the program finds a >position in the EGTBs it doesn't have to look at all the transpositions anymore >that can arise from this position, so even if NPS goes way down that would not >be telling you everything. Anyway, don't really know all that much about this >stuff, how important this transposition effect is in relation to movegeneration >for example with fewer pieces. I just thought move generation couldn't really >have that much effect on NPS? > > >I see now that José Carlos also mentioned the effect of transpositions but that >Robert Hyatt said that he doesn't see how that should affect the NPS. I am >afraid I don't really understand that unless, Robert, you don't count in the >hashtable hits in the number of positions evaluated? > >Regards, Eelco That is the problem. I call Search(), and right at the top, do a hash probe. If I "hit" then I return the score and yes, the node is counted. But all the search overhead was still done, with no moves searched. High hash hits tends to make your NPS drop, but makes your depth go up...
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.