Author: Vincent Diepeveen
Date: 09:33:47 09/23/03
Go up one level in this thread
On September 23, 2003 at 12:14:15, Vincent Diepeveen wrote: I need to make one note. All tests published at aceshardware.com are 32 bits tests. So it's tested under the latest windows version which is 32 bits. Simply compiled at home executables and shipped those to get tested. However i also tested at the quad opteron in 64 bits mode. This was a big bummer. First of all it was pretty fast. quad opteron 1.8Ghz got 430k nps. That was for a special NUMA version of DIEP. however the linux kernel is having major bugs scheduling, so we must wait for them to get fixed, for better results. It's a good start though. For the dual Xeon other versions were used which were SMP versions. Regrettably the SMP version no longer compiles very well under linux, so i could not compare it as i had only 1.5 days access to the quad opteron. 4 processes was 227k nps at the dual Xeon. I was very dissappointed by gcc in 64 bits. Johan tried the 32 bits windows smp version and got 404k nps with it at the quad opteron. GCC is doing so very poor that i don't even need to comment at it. Single cpu a 32 bits version is simply faster. It's that i managed to get 5% extra with PGO out of it, otherwise we wouldn't know. The tested executables at aceshardware are all default msvc compiled versions version 6.0 sp5 procpack. Intel c++ isn't faster than that at the P4, it's a lot slower in fact (without PGO), more than 1.5%. Nevertheless, seeing the speed of the opteron/amd64 in 32 bits under windows i can't take away my positive surprise. Just imagine when we get 16 registers for it soon too! Best regards, Vincent
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.