Author: Bryan Hofmann
Date: 18:47:06 02/22/05
Go up one level in this thread
On February 21, 2005 at 23:45:20, Robert Hyatt wrote: >On February 21, 2005 at 21:08:05, Bryan Hofmann wrote: > >>On February 20, 2005 at 19:56:51, Peter Skinner wrote: >> >>>On February 20, 2005 at 14:48:24, mike schoonover wrote: >>> >>>>hi all, >>>>been noticing this problem for a while with crafty. >>>>exits in ics mode quite freaquently. >>>>more with the newer ones. >>>>see:http://wbforum.volker-pittlik.name/viewtopic.php?t=1680 >>>>it is not compile pessific. >>>>just wondering,is this a crafty or wb problem. >>>>help appreciated. >>>>regards >>>>mike >>> >>>I read your post on the wb forums, and the replies by Bryan Hoffman. >>> >>>I decided to test his version vs mine on the same computer, as I haven't done it >>>in a while. Here are the results: >>>-------------------------------------------------------------------------------- >>>unable to open book file [./book.bin]. >>>book is disabled >>>unable to open book file [./books.bin]. >>>hash table memory = 24M bytes. >>>pawn hash table memory = 6M bytes. >>> >>>Crafty v19.19 BH >>> >>>White(1): bench >>>Running benchmark. . . >>>...... >>>Total nodes: 89729038 >>>Raw nodes per second: 766914 >>>Total elapsed time: 117 >>>SMP time-to-ply measurement: 5.470085 >>>White(1): >>> >>>EPD Kit revision date: 1996.04.21 >>>unable to open book file [./book.bin]. >>>book is disabled >>>unable to open book file [./books.bin]. >>>hash table memory = 24M bytes. >>>pawn hash table memory = 6M bytes. >>> >>>Crafty v19.19 (1 cpus) >>> >>>White(1): bench >>>Running benchmark. . . >>>...... >>>Total nodes: 89729038 >>>Raw nodes per second: 787096 >>>Total elapsed time: 114 >>>SMP time-to-ply measurement: 5.614035 >>>White(1): >>>------------------------------------------------------------------------------- >>> >>>Mine still seems slightly faster, and is the default compile with VC++ 2005 >>>Express. >>> >>>Using these options: >>> >>>cl /Ox /O2 /GL /Gs /GA /GF /GT /Gr /MT /w /DNT_i386 /DWIN32 /D_CONSOLE /DWINDOWS >>>/DFAST /DEGTB6 /DEPD /DFUTILITY /DVC_INLINE_ASM crafty.obj egtb.obj >>> >>>Peter >> >>I find this hard to believe as I just ran your compile vs my compile on two >>different systems and Intel Pentium 4 2.8 GHz and a AMD 3000+ XP system. I'm >>using the full VC 2005 with POGO. The largest difference is in the AMD and I >>sure this is due to the POGO is being done on the AMD system. >> >> >>AMD 3000+ XP >> >>Skinners >>Crafty v19.19 >> >>White(1): ben >>Running benchmark. . . >>...... >>Total nodes: 96761642 >>Raw nodes per second: 1256644 >>Total elapsed time: 77 >>SMP time-to-ply measurement: 8.311688 >>White(1): quit >> >>Crafty v19.19 BH >> >>White(1): ben >>Running benchmark. . . >>...... >>Total nodes: 96761642 >>Raw nodes per second: 1362840 >>Total elapsed time: 71 >>SMP time-to-ply measurement: 9.014085 >> >> >> >> >>Pentium 4 2.8GHz >> >>Skinners >>EPD Kit revision date: 1996.04.21 >>unable to open book file [./book.bin]. >>book is disabled >>unable to open book file [./books.bin] >> >>Crafty v19.19 >> >>White(1): ben >>Running benchmark. . . >>...... >>Total nodes: 96761642 >>Raw nodes per second: 1018543 >>Total elapsed time: 95 >>SMP time-to-ply measurement: 6.736842 >>White(1): quit >> >>Mine >>unable to open book file [./book.bin]. >>book is disabled >>unable to open book file [./books.bin] >> >>Crafty v19.19 BH >> >>White(1): ben >>Running benchmark. . . >>...... >>Total nodes: 96761642 >>Raw nodes per second: 1063314 >>Total elapsed time: 91 >>SMP time-to-ply measurement: 7.032967 >>White(1): > > >Hate to tell you guys, but you are all pissin' in the wind. :) Doubtful but you are allowed to have your view. > >optimizations change from one processor to another, and I am not just talking >about AMD vs Intel. Different memory timing, different cache size/timing, >different memory latency, different processor timings, the list goes on and on, >and each can affect the speed of the program sporadically and unpredictably. >Even poor memory/cache aliasing can make the same executable vary in speed >significantly from one day to the next on the same processor. > >Benchmarking and optimizing is not a "compile one time, run one test, and look >at the results". It is a "compile once, run a bunch of tests, then clear memory >and run the same thing again. Multiple times... Whether you average or use the >best/worst/typical result is up to you, but there are too many variables for one >person to compile and think "this is the best there is". There are even >compiler and optimizer differences to contend with beyond hardware difference... This is exactly what I have done time and time again with my compiles. I take it a step futher in that I have forced inlined some functions to attain a quicker compile and tested and both Intel and AMD platforms and had others test the compiles. All have pointed to one thing, the compiles I produce are faster then any of the others out there on a windows platform.
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.