Author: Vincent Diepeveen
Date: 12:41:31 08/03/03
Go up one level in this thread
On August 02, 2003 at 13:09:13, William Penn wrote: With current DIEP i get: all is SINGLE cpu with 400MB hashtables: P4 3.06Ghz : 95k nps (visual c++ 6.0 sp5 procpack) K7 2.127Ghz (MP2600) : 100k nps a bit more than that with gcc 3.3 R14000 500Mhz : 26k nps MIPSPRO compiler Madison 1.3Ghz 3MB L3: 62k nps (gcc 2.96). Intel c++ 7.0 not working parallel yet at machine (i cannot find working functions that will lock volatile int in shared memory for DIEP for interprocess communication in openmp), but single cpu i have gotten madison to 74-76k nps already when using stuff like: # intel c++ nu CC = ecc CPP = ecc #CFLAGS = -g -DUNIXPII #CFLAGS = -O3 -tpp2 -mcpu=itanium2 -openmp -prof_genx -DUNIXPII -DIA64 -DCONFIG _SMP CFLAGS = -O3 -tpp2 -mcpu=itanium2 -openmp -prof_use -DUNIXPII -DIA64 -DCONFIG _SMP So that is very dissappointing performance for intel c++. With this the difference itanium versus k7 is: 1.28 to 1. Better than any other cpu, but very poor considering the theoretical potential of the Itanium. The problem of the itanium is the software trivially. It executes 2 bundles of 3 instructions each clock. After i strip executable then: -rwx------ 1 vdiep chess 1842312 Aug 3 21:40 diep 1.8MB for executable at itanium! Must be NOP NOP NOP. Note that this is gcc executable. intel executable even worse. >On August 01, 2003 at 02:42:23, Derek Paquette wrote: > >>My CPU >>Athlon 1700xp >>128 RAM *in hash* out of 256 >> >>*all node tests are done at board reset, infinite analysis >>Shredder 7.04 : >>I am reaching about 230 to 260 kn/sec >> >>Fritz 8 >>I am reaching about 550 to 600 kn/sec >> >>Crafty 19.03 >>I am reaching about 500 kn/sec >> >>What kinds of processors do you have, and what sort of speeds are you getting, >>please only reply if you have the same engines, if not, i cannot compare >> >>I am THINKING of upgrading to a Athlon 2800XP >>how much of a speedup would I expect? >>thanks > >I'm running W98se, 256MB, 500MHz Celeron processor. Using the Shredder 7 (CB) >GUI and Shredder 7.04 UCI engine I typically get about 70-80 kN/s in the >opening/middlegame. In simpler endgames it may go over 100 kN/s, perhaps as high >as 130 kN/s. > >Interestingly I can get much faster kN/s speeds for this engine if I use the >Shredder Classic 6 GUI instead. I believe it has something to do with the hash. >The hash table takes a long time to initialize (1-2 minutes) with the Shredder 7 >(CB) GUI but is essentially instantaneous with the Shredder Classic 6 GUI. >Perhaps that doesn't matter much with longer analyses of an hour or more(?). > >In most cases I'm using 160MB hash which experience has taught me works pretty >well on my system, allowing normal multitasking of other apps at the same time. >The GUI says I can use 190MB, which I do occasionally, but it slows down >multitasking considerably. Like you I use infinite analysis mode. > >Also like you, I'm thinking about upgrading soon, and wonder which processor(s) >would give me the most advantage...? >WP
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.