Author: Albert Bertilsson
Date: 05:15:34 12/12/03
Go up one level in this thread
On December 12, 2003 at 07:24:47, Ingo Bauer wrote: >Hi > >The next thing would be to see the position (at ply 3) that the comps are >crunching on at the moment :-) (but pls no movement for each calculated pos :-) >) If we attempt to calculate perft(12) I'll include that to the client. I already have most of the code in some bits. > >Because you have the KNPS now you can either ask if someone is entering >information about the processor/speed/hash (or identify these three values by >the client) and you would have a nice database about relativ processor-speeds >itself. > >With a database big enough you could see when the pocessors reach the point >where it is sensless to add more hash! For perft(8) your computer showed that going from 512 to 1024 gives an extra 8% in performance, I guess that a few GB of hash is where the performance gain will be very little. > >... > >Bye Ingo > > >>I've collected some info posted in the distributed perft thread. >> >>Using Sharper 0.17p to do perft calculations with hash table has given the >>following results: >> >><PRE> >>Hardware Hash CS Nodes MNPS >>Athlon XP (Barton) at real 2400MHz 256 51254 84998978956 165 >>P4 3.06Ghz 256 49801 84998978956 170 >>Athlon XP (Barton) at real 2400MHz 512 45422 84998978956 187 >>Athlon XP 2.5GHz, 200fsb(400DDR) with 3-4-4-10 512 43848 84998978956 193 >>Athlon XP (Barton) at real 2400MHz 1024 42027 84998978956 202 >></pre> >> >>Single cpu machines counting 202 million nodes per second is really impressive. >> >>If you have a machine that you'd like to bench, download the distributed perft >>client (Sharper 0.17p is included with it). Set the hash size with "hashsize >>XXX" (XXX = number of megabytes of ram) and run "perfthash 8" and report the >>time it took to finnish the calculation. >> >>I'm going to add some lower spec. machines later. >> >>/Regards Albert
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.