Linux client slower?

I'm using proth_sieve_040_win_cmov on an Athlon XP 2100+ (Windows 2000 SP4) and proth_sieve_040_linux_cmov_static on an Athlon XP 2500+ (Fedora R1, Custom 2.6.0 Kernel).
Even though the linux computer is 100MHz (1837) faster with a 256KB larger cache, its seiving rate is equal to or less than the the windows computer (1737). Both clients show the output rate at 524kp/s under optimal conditions, and in the stats.txt the linux client show about 502kp/s and the windows client shows 507kps.

I have both boxes running a minimal configuration, with no unnecessary processes running.

Would factors such as CAS latency on the ram make a difference, or is this software/OS based.

I'll do more testing on this tomorrow and see if i can find any differences.

Thanks for any information.
pixl97

UPDATE!!!

It does appear with my setup that the linux client (both static and non static) is slower than the windows client in both the 2.4 and 2.6 kernel (glibc-2.3.2-101.1)

On the same computer...
Windows rate (559kp/s)
Linux rate (524kp/s)

Is this a compiler optimization difference? btw what compiler do you use for the linux client?

Hi,

memory latency and things like that no doubt makes a difference, yes. The difference between OSs on the same machine is due to, as you say, compiler optimization differences and to some extent OS differences.

g++ v3.3.2 is used to compile the linux client.

Just wondering, have the developers (MikeH?) tried to use this...

http://www.intel.com/software/produc...lin/noncom.htm

to see if there was a performance difference in the resulting code? I've heard of many cases where the resulting intel code is faster than the gcc code?

Just so you know, I have been called a performance junky before. :)

The developers would be me and Paul.

Your link looks interesting... I was under the impression that you could no longer download ICC-Linux for free, but it seems I was wrong. I'll have a look.

Mikael
- binary craftsman.