Author: Tony Werten
Date: 15:27:05 11/09/01
Go up one level in this thread
On November 09, 2001 at 14:07:57, Robert Hyatt wrote: >On November 09, 2001 at 09:44:14, Tony Werten wrote: > >>On November 09, 2001 at 09:40:29, Robert Hyatt wrote: >> >>>On November 09, 2001 at 02:16:49, Tony Werten wrote: >>> >>>>On November 08, 2001 at 18:04:43, Robert Hyatt wrote: >>>> >>>>>On November 08, 2001 at 14:46:15, Tony Werten wrote: >>>>> >>>>>>On November 08, 2001 at 12:01:16, Robert Hyatt wrote: >>>>>> >>>>>>>On November 08, 2001 at 07:45:59, Vincent Diepeveen wrote: >>>>>>> >>>>>>>>On November 07, 2001 at 15:51:59, Robert Hyatt wrote: >>>>>>>> >>>>>>>>>On November 07, 2001 at 12:22:05, William Dozier wrote: >>>>>>>>> >>>>>>>>>>Good day to all of you, Question: Is there a killer book for black? Is there a >>>>>>>>>>book for black, so it can counter all the white moves. When does the table base >>>>>>>>>>start kicking in; in crafty? >>>>>>>>> >>>>>>>>>I doubt that there is any "killer book" for black or white, assuming your >>>>>>>>>opponent does his homework. >>>>>>>>> >>>>>>>>>tablebases start getting hit when there are around 16 pieces left on the board, >>>>>>>>>most of the time. Others might get different results of course. I have seen >>>>>>>>>hits with 20 pieces left... but not very often. >>>>>>>> >>>>>>>>With all extensions turned on i get hits from openings position >>>>>>>>already after a night of search with big hashtable. >>>>>>> >>>>>>>Overnight that is possible. I interpreted it as "during a real game". :) >>>>>> >>>>>>Depends on where you probe. I probe everywhere and have seen tablebasehits with >>>>>>30 pieces on the board in 3 minute searches. Useless but funny. >>>>>> >>>>>>Tony >>>>> >>>>> >>>>>What happens when you get into 8-10 piece endings? Does your speed die? I >>>>>did that early on and actually lost winnable games, because I slowed down so >>>>>bad the search depth was not enough to reach won positions. IE from 100K it >>>>>would drop to 1-5K... >>>> >>>>If I have a 10 ply search I will probe the first 10 ply on harddisk, for the >>>>rest I use my own tablebaseformat wich stores the 4 and 3 pieces in memory, >>>>making probing not more expensive than an evaluation. >>>> >>>>At the dutch open I set up a testposition with 6 pieces and still got 1Mn/s on a >>>>AMD 1700+ >>>> >>>> >>>>Tony >>>\ >>>That isn't "probing everywhere". IE with a 10 ply search, you can extend to >>>way beyond 20 plies. Probing out there will hurt. And if you probe in the >>>q-search it will hurt even more. >>> >>>Your limit is the same as I have been using for years... >> >>Sorry, probing the 5 pieces the first 10 ply and probing 4 pieces in the rest of >>the tree is not probing everywhere ? > > >No. I interpret the question as "probing endgame tables on disk". The 3-4 >piece files don't hurt a thing. But probe the 5-6 piece files in the q-search >and things are different. Aha, now I see. Nope, not any luck yet. I guess with striping and a big buffer you can probe from disk the complete full search (so including extensions ) but qsearch probing is a bit too much. Having a big hash might help as well, since you don't reach 8 pieces suddenly most of the time. But when you do after a big exchange sequence it kills you. Setting up a test position as well. > > >> >>Don't get it. >>I don't probe 6 pieces, does that mean I probe nowhere ? >> >>Tony > >"probe" in this context was refering to disk I/O. That is what kills >performance, not poking around in memory. Been a while since I have poke'd in memory. I think it was commodore 64. (And the computers we had in highschool. Don't remember what it was, do know that the upperleft corner of the screen was 33792.) Tony
This page took 0.01 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.