Author: Vincent Lejeune
Date: 04:47:06 05/11/02
Go up one level in this thread
On May 11, 2002 at 06:15:51, Joshua Lee wrote: >I have been working on the various test sets at 90 minutes per position for many >months now and have gone from over 1700 to now 1438 >it may not seem like i have done much but when you consider i started with >several thousand and that i have 512 Not Solved in 90 minutes >and many Solved that took upwards of 80+minutes to solve...... > >CCR.07 - 5033s / 17 find Rfb8 >[D] r4rk1/3nppbp/bq1p1np1/2pP4/8/2N2NPP/PP2PPB1/R1BQR1K1 b - - 0 1 > >I have a very interesting question: If it takes this long and longer to solve >these problems and considering the rating on a 1Ghz system, What would the Elo >rating be on lets say 27Ghz? Obviously many problems left over ECE/MES/ECM are >full of errors but how many of these that are solveable will take over 5 hours >to solve? If you talk about new 8*3400+ opteron system the speed for chess have to be computed with the Bob's formula : 3400+3400*70%*7 =~ 16500 "usefull" Mhz 1) test solving : about 16 times speedup if you use parallelle algo or 27* (may be less if you count memory bottleneck) if use 8 instances (one on each CPU) to solve 8 differents positions in the same time 2) rating against comp : if 2* speedup means +60 elo, 16* means +240 elo !!!! 3) rating against human : very hard to say, playing against strong humans is more a matter of knowledge than a matter of speed (however it means about 3 plies deeper, very usefull where there is no deep plan to understand or in a complicated tactical position)
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.