Author: Robert Hyatt
Date: 13:50:06 09/06/99
Go up one level in this thread
On September 06, 1999 at 15:22:29, Jeff Anderson wrote: >Can someone please explain to me how a problem with computer processor can >cause it to play a move that is would not prefer at any ply on otherwise normal >hardware? I could understand how a problem with the processor would cause it to >not search deep enough, or not search at all, and therefore play a weaker move >than it would have, if it had performed a satisfactory search; But how can a >problem with the processor cause software like Rebel to play a move that it >would never prefer at all if it had been running on proper hardware? >Thanks, >Jeff This happens on many computers when they fail. The failure can be one of many things... The whole machine can just hang, or it can forget how to do one thing properly. An example: Several years we ran on a prototype (serial number 1) of the Cray C90. It had not been 'burned in' for the usual 30 days before we started to use it in an ACM event. And we started seeing some funny things. Turns out that exactly 1 cpu developed a problem with integer multiply. If you multiplied two numbers you _always_ got a positive number. Worked _wonders_ for my evaluation as you might guess. I have seen several log files of crafty on an overclocked AMD where it simply produced impossible scores, impossible moves, etc. The problem is that the internal data paths are 32 bits wide, and each bit passes thru several gates between two clock cycles. One slightly slow gate can make one bit wrong every 1,000,000 cycles. That is all that is needed to screw up a program.
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.