Author: Dann Corbit
Date: 22:11:53 05/11/00
Go up one level in this thread
On May 12, 2000 at 01:08:40, Jouni Uski wrote: >On May 12, 2000 at 01:02:43, Dann Corbit wrote: > >>On May 12, 2000 at 01:00:14, Jouni Uski wrote: >> >>>I have now played 146 blitz games between Lg2000v2.7 and Crafty17.10. I have >>>used both AMD450 and Celeron300 (no ponder, because one PC at time). Different >>>books are used Nunn1, General.ctg and also their own books. Totally lg2000 >>>leads 75,5 - 70,5. Level 4m+1s or 60/5m. >>> >>>I noticed, that in this position from correspondence challenge: >>>[D]r6k/7p/5pp1/7n/2bQN3/5q2/P1P4R/K2R4 w - - 0 0 >>> >>>LG2000 reach 1113knps in AMD450 and this with 8 MB hash. With bigger hash >>>nps goes lower, but is still over 1000k - is this normal behaviour? >> >>The nodes calculated may be less because they were already found in the hash >>table (calculation skipped for those nodes). How long did you let it run and >>what was the time to reach ten ply under (for instance) 8M verses 64M? > >Here is comparison with: 8MB 16MB 50MB > >time to 12 ply 5'04'' 4'38'' 4'11'' >nps after 10 min 1113k 1036k 1001k This makes the explanation clear. Since it is making the plies faster with bigger hash, it is clear that the higher NPS figures mean more hash table lookup misses meaning it has to calculate the value over again. Another effect I have seen like this one is when endgame tablebase files are installed. I saw the NPS figures for crafty drop dramatically for some endgame EPD positions. But the time to ply dropped even more.
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.