Author: Carlos del Cacho
Date: 11:04:15 06/20/01
Go up one level in this thread
On June 20, 2001 at 13:44:23, Dann Corbit wrote: >On June 20, 2001 at 13:14:37, Carlos del Cacho wrote: > >>On June 20, 2001 at 07:35:24, Uri Blass wrote: >> >>>Crafty needs only 2.2 seconds to calculate perft 5(the number of legal games >>>with 5 moves) and only 54.76 seconds to calculate perft 6(119,060,324 nodes) >>> >>>It is clear that Crafty does not use hash tables to calculate perft faster. >>> >>>Is there another program (with free or without free source code) that can >>>calculate perft (without hash) faster than Crafty? >>> >>>Uri >> >> >>It is a well known fact that bitboards are slow at move generation using 32 bit >>architecture hardware. Some results on my K6 400 MHz >> >>Crafty 17.14 | Pepito | A 0x88 thing I'm messing with (ANSI C) >> | | >>ply 5: 8.02 | ply 5: 6.91 | ply 5: 4.17 >>ply 6: 196.96 | ply 6: 161.00 | ply 6: 90.50 > >Depends on the chip used. Here's an AMD 950 MHz: >--------------------------------- >Crafty v18.10 > >White(1): perft 5 >total moves=4865609 time=1.97 >White(1): >--------------------------------- >Pepito> perft 5 > >Time spent: 2.07 >Paths searched: 4865609 >--------------------------------- Oops, this was my most recent binary. For sure it's even slower than that on the Athlon (and I could create a faster binary of Crafty for the K6). But what I mean is that 2 instructions per 1 is too much a handicap for bitboards if your intention is to go fast (on current hardware, by current I mean affordable :-). Carlos
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.