mersenneforum.org  

Go Back   mersenneforum.org > Factoring Projects > Factoring

Reply
 
Thread Tools
Old 2010-02-26, 10:10   #1
siew
 
Oct 2009

1111112 Posts
Default Choosing the best CPU for sieving

Hi all,

My question is: what the better choice for CPU to reach maximum perfomance when running GGNFS sieving?

I tested some CPUs already, but seems there some questions is open.

equal task - seiving the 155 digit number.

Core I7-920 - 0.45-0.47 rel/sec
Core I5-540 - 0.69-0.71 rel/sec
Core2 Q6600 - 0.49 - 0.52 rel/sec

What params can affect to speed? CPU cache, bus clock or anything else?
How to choose the best cpu for computing this task?
Tanx!
siew is offline   Reply With Quote
Old 2010-02-26, 10:35   #2
ET_
Banned
 
ET_'s Avatar
 
"Luigi"
Aug 2002
Team Italia

12CD16 Posts
Default

Quote:
Originally Posted by siew View Post
Hi all,

My question is: what the better choice for CPU to reach maximum perfomance when running GGNFS sieving?

I tested some CPUs already, but seems there some questions is open.

equal task - seiving the 155 digit number.

Core I7-920 - 0.45-0.47 rel/sec
Core I5-540 - 0.69-0.71 rel/sec
Core2 Q6600 - 0.49 - 0.52 rel/sec

What params can affect to speed? CPU cache, bus clock or anything else?
How to choose the best cpu for computing this task?
Tanx!
If you work with multuthreading (more than 1 core) i5 and i7 have a better handling of memory transfer.

Luigi
ET_ is offline   Reply With Quote
Old 2010-02-26, 10:38   #3
siew
 
Oct 2009

32×7 Posts
Default

I working using
gnfs-lasieve4I??e.exe

with many threads

so wants to improve the speed by choosing right CPU
siew is offline   Reply With Quote
Old 2010-02-26, 11:32   #4
ET_
Banned
 
ET_'s Avatar
 
"Luigi"
Aug 2002
Team Italia

12CD16 Posts
Default

Quote:
Originally Posted by siew View Post
I working using
gnfs-lasieve4I??e.exe

with many threads

so wants to improve the speed by choosing right CPU
If you have no price limitations, the core i7 is the best choice for you.

Luigi
ET_ is offline   Reply With Quote
Old 2010-02-26, 15:29   #5
henryzz
Just call me Henry
 
henryzz's Avatar
 
"David"
Sep 2007
Cambridge (GMT/BST)

5,869 Posts
Default

an Athlon II 620 x4 is a little faster and probably cheaper than a Q6600
henryzz is offline   Reply With Quote
Old 2010-02-26, 15:52   #6
Mini-Geek
Account Deleted
 
Mini-Geek's Avatar
 
"Tim Sorbera"
Aug 2006
San Antonio, TX USA

426710 Posts
Default

I wouldn't guess a quad-core i5 would be any slower than an i7, and would certainly be cheaper.
Mini-Geek is offline   Reply With Quote
Old 2010-02-26, 15:57   #7
fivemack
(loop (#_fork))
 
fivemack's Avatar
 
Feb 2006
Cambridge, England

11000111011112 Posts
Default

Could you confirm that these timings are for individual sievers when running one siever per core/thread of the machine (IE eight on i7/920, four on i3/540, four on Q6600) ?

I found that eight sievers on an i7/920 each ran a bit slower than individual sievers running four on a Q6600, but since you had twice as many the aggregate performance was significantly better.

Phenom and Q6600 are about the same speed for me; my Opteron 2380 is about 10% slower than the same-MHz Phenom 9850, but I think that's to be expected with memory access on a two-socket motherboard.
fivemack is offline   Reply With Quote
Old 2010-02-26, 15:59   #8
fivemack
(loop (#_fork))
 
fivemack's Avatar
 
Feb 2006
Cambridge, England

13×491 Posts
Default

May I suggest that you run the benchmark in

http://www.mersenneforum.org/showthread.php?t=10003

on the various machines?
fivemack is offline   Reply With Quote
Old 2010-02-26, 18:40   #9
siew
 
Oct 2009

1111112 Posts
Default

Code:
ggnfs-svn374-win32-p4\test1>gnfs-lasieve4I14e -a 1188.poly -f 50000000 -c 1000 -o bench -v
Core I5 430M - 0.287 sec/rel
siew is offline   Reply With Quote
Old 2010-02-26, 18:45   #10
siew
 
Oct 2009

32×7 Posts
Default

Quote:
Originally Posted by fivemack View Post
Could you confirm that these timings are for individual sievers when running one siever per core/thread of the machine (IE eight on i7/920, four on i3/540, four on Q6600) ?
Yes, the timings is for one siever on one thread, i.e. each siever has such timing on it's core.
siew is offline   Reply With Quote
Old 2010-02-26, 19:31   #11
Batalov
 
Batalov's Avatar
 
"Serge"
Mar 2008
Phi(4,2^7658614+1)/2

3·43·73 Posts
Default

...and you are using a 32-bit windows binary, right? (you mentioned .exe)

A proper 64-bit Linux binary will be almost two times faster on any of them, for the same money. A 64-bit Windows binary doesn't yet exist. On AMD CPUs, there may be an additional small speed-up if you re-compile to use all of its 64Kb/core L1 cache (all Intel cores have 32Kb/core), but this is far from a decision maker. You may want to decide on a benchmark set, run it, and derive a relations/c/$ (and somehow include electric costs, maybe).
Batalov is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Unhappy in choosing prime sequence :-) pepi37 Lounge 9 2017-07-15 19:53
Choosing the best polynomial wombatman Msieve 123 2013-08-27 11:27
Help choosing motherboard please. Flatlander GPU Computing 4 2011-01-26 08:15
MPQS: choosing a good polynomial ThiloHarich Factoring 4 2006-09-05 07:51
Choosing amount of memory azhad Software 2 2004-10-16 16:41

All times are UTC. The time now is 20:19.

Fri May 7 20:19:23 UTC 2021 up 29 days, 15 hrs, 1 user, load averages: 2.98, 2.79, 2.57

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.