oops, as the list admin, I commited the cardinal sin.  While dumping the 
morning load of 10-20 spams, I accidentally hit 'discard' on this 
message...


Subject:
Measures with PThread GLucas on new 2-nodes x 2-chips x 2-cores 
Itaniume2 machine
From:
[EMAIL PROTECTED]
Date:
Fri, 22 Sep 2006 17:05:21 +0200

To:
The Great Internet Mersenne Prime Search list <[email protected]>
CC:
[EMAIL PROTECTED]


Hello,

Look at: http://www.mersenneforum.org/showthread.php?p=87553#post87553

I've posted results of measures I've done with GLucas multi-threaded on this new
Itanium2 machine.

Surprisingly:
- running GLucas multi-threaded is only -1 to +4 % slower than GLucas with no
threads,
- GLucas runs faster with 4 threads than with 1, 2 or 3.
(I mean:
     let the sec/iter value measured with N threads is: SI_N
     let the sec/iter value measured with 1 thread  is: SI_1
     let the sec/iter value measured with 0 thread  is: SI_0
then the measures I've done show:
         SI_1 / (SI_N * N)  is 120 % for N=4

I've done the measures several times. Though each point in the figure is not
perfectly stable, I always have the same curve.

Do not see where I could be wrong ...

Regards,

Tony

_______________________________________________
Prime mailing list
[email protected]
http://hogranch.com/mailman/listinfo/prime

Reply via email to