Author: Vincent Diepeveen
Date: 19:44:57 06/13/03
Go up one level in this thread
On June 13, 2003 at 19:28:34, Brian Richardson wrote: Hello your opteron performance is not correct. If you native compile 64 bits with GCC at the opteron crafty with branch optimizations (that means you compile 1 time with profile option then run it for a number of minutes then recompile, gives big extra speed) and then run dual you should get about 4 million nodes a second. And not a node less than that. Even at specint the 64 bits gcc compiled version is about 60% faster than bob's own machine has. specint is not using even inline assembly nor profile reinforced code (which matters really a lot at the opteron and near to zero at the itanium2 becuase it's a buggy predication processor). Best regards, Vincent >Note the time differences between versions, and the relative SMP scaling. > > Version CPUs Total SMP Raw nodes > elapsed time-to-ply per second > time >Opteron 244 (1.7GHz) 19.03 2 28 22.857143 1976833 >Opteron 244 (1.7GHz) 19.03 1 43 14.883721 1234812 >AMD 1900+ (1.6GHz) 19.03 2 39 16.410256 1408076 >AMD 1900+ (1.6GHz) 19.03 1 56 11.428571 948159 >Opteron 244 (1.7GHz) 18.15 2 48 13.333333 1807684 >Opteron 244 (1.7GHz) 18.15 1 77 8.311688 1097373 >AMD 1900+ (1.6GHz) 18.15 2 71 9.014085 1278866 >AMD 1900+ (1.6GHz) 18.15 1 104 6.153846 812478 >Itanium2 1GHz 18.15 2 112 5.714286 959705 >Itanium2 1GHz 18.15 1 186 3.44086 524126 > >Note all runs with: >hash table memory = 384M bytes. >pawn hash table memory = 24M bytes. >EGTB cache memory = 1M bytes.
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.