Author: Robert Hyatt
Date: 07:17:59 10/13/99
Go up one level in this thread
On October 13, 1999 at 08:54:43, Jouni Uski wrote: >On October 13, 1999 at 08:09:13, Robert Hyatt wrote: > >>On October 13, 1999 at 01:10:30, Harald Faber wrote: >> >>>On October 13, 1999 at 01:04:05, Jouni Uski wrote: >>> >>>>Yes from latest SSDF list You see that Hiarcs7.32 with tablebases is only >>>>2 points over Hiarcs7 without them. Hmm. Not much benefit from 100+ MB packed >>>>data... >>>> >>>>Jouni >>> >>>Argh!!! >>>The question is how much below would Hiarcs7.32 be WITHOUT the tablebases... >> >> >>Even more important, how would _other_ programs do? Hiarcs has an apparently >>very conservative approach to probing in the search, based on some numbers that >>were posted here recently. It is not surprising that tablebases don't make any >>difference given that circumstance. > >Bob. I just installed Crafty 16.15 (TB version) under Fritz5.32. It's >unbelievable how Crafty can access TB dynamically (in search) without slowing >NPS much. >How is this possible? I remember when I tested same thing with Mchess 7 NPS >go down to about 1knps (it was P90 PC then). > >Jouni Several things. (1) Eugene's buffering code is quite good; (2) if you only probe _after_ a capture, and only when pieces <= 5, the probes aren't done a ridiculous number of times. (3) I did a _lot_ of testing to find a way to steadily increase the depth at which probes occur without smashing the search back to 1K nps as you mentioned. And don't forget that a CD will kill me. And if you really want to see it probe, try 10K rpm LVDS drives at 80 mbytes/second transfer rate.
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.