mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2010-02-21, 17:54   #1
SO7783
 
Nov 2003
77377/77477 (either way)

22×3 Posts
Default Running 50% util on i5 and Core2 (same MHz)

I'm running the same 7 month old version of Prime95 on two Intel boxes that should be very close in effective output on double checking using 1280K FFT in the 222xxxxx range.

The Core2 Duo uses one thread of 2 physical cores and the i5 is two hyperthreaded cores so I set it for a single worker window and 2 multithreaded "cores" so both machines are at fifty percent utilization.

So why does the newer i5 (same true cores, same clockspeed) burn through tests in half the time using the same resources? Looking through RESULTS.TXT it is pretty obvious my lower-end Intel i5 gets almost identical results running at two threads of 50% or four threads of 100%

thanks for any clarification!
SO7783 is offline   Reply With Quote
Old 2010-02-21, 19:24   #2
axn
 
axn's Avatar
 
Jun 2003

23×607 Posts
Default

Quote:
Originally Posted by SO7783 View Post
so both machines are at fifty percent utilization.
What do you mean by this? Your Core 2 Duo should be showing 100% if P95 is properly configured. As it is, it is only using half its capacity.

Core i5, since it is hyperthreaded, is ok even if it is only showing 50%. It is in fact using full capacity.

Last fiddled with by axn on 2010-02-21 at 19:24
axn is online now   Reply With Quote
Old 2010-02-22, 04:51   #3
SO7783
 
Nov 2003
77377/77477 (either way)

22×3 Posts
Default

I simply want to run both Intel-based computers at 50% ... not 100%

When I run the core2duo and the i5 in single benchmark mode they are virtually indistinguishable. When I run the built-in benchmark for the two machines using both processors fully they are nearly identical.

But the i5 running doublechecks at half-CPU gets twice the work accomplished as the Core2 at half the total CPU capabilities. You see it instantly in the per iteration times while the WIN7 task manager remains at 50% usage on both machines.

Where's the free lunch coming from with the i5 650 (two cores each with two hyperthreads) that it suddenly can run twice the speed under one specific set of test conditions?
SO7783 is offline   Reply With Quote
Old 2010-02-22, 07:22   #4
sdbardwick
 
sdbardwick's Avatar
 
Aug 2002
North San Diego County

10101010102 Posts
Default

The i5 is probably running each thread on its own physical core. If you want to run it at 50% cpu utilization, run 1 worker with 1 thread; due to Prime95's highly optimized assembly language routines, using the virtual (hyperthreaded) core with Prime95 generally has little positive or negative effect.

If you are dead set on using 1 physical and 1 virtual core, you will need to experiment with the AffinityScramble setting.
sdbardwick is offline   Reply With Quote
Old 2010-02-22, 18:00   #5
SO7783
 
Nov 2003
77377/77477 (either way)

C16 Posts
Default

I'm not fighting the program's intent; it is more a case of wanting to test one number at a time, while keeping temperatures and power usage in a reasonable range.

This is my 12th year using P95 regularly and I've been putting dual processor machines through mild stress tests since the Pentium Pro 180 and PII-300 days. Plus, I do have other processes running on my computers

Most everything else running in the background does look to the Windows CPU figure for system utilization -- and they behave differently when they see 100% rather than half.

Unfortunately I am still not satisfied with any explanation why on this ONE TEST that 2 nearly identical machines are suddenly off by a factor of two. (My machines are not very slow, but apparently this week I am)
SO7783 is offline   Reply With Quote
Old 2010-02-22, 20:47   #6
sdbardwick
 
sdbardwick's Avatar
 
Aug 2002
North San Diego County

2·11·31 Posts
Default

Quote:
Originally Posted by SO7783 View Post
So why does the newer i5 (same true cores, same clockspeed) burn through tests in half the time using the same resources? Looking through RESULTS.TXT it is pretty obvious my lower-end Intel i5 gets almost identical results running at two threads of 50% or four threads of 100%
They are not using the same resources. C2D is using 100% of one physical processor, i5 is using 100% of two physical processors. Hyperthreading presents in Windows as two additional processors, so it reports 50% [(100+100+0+0)/4] overall utilization.

Last fiddled with by sdbardwick on 2010-02-22 at 20:48
sdbardwick is offline   Reply With Quote
Old 2010-02-23, 15:54   #7
joblack
 
joblack's Avatar
 
Oct 2008
n00bville

10110101012 Posts
Default

Quote:
Originally Posted by axn View Post
What do you mean by this? Your Core 2 Duo should be showing 100% if P95 is properly configured. As it is, it is only using half its capacity.

Core i5, since it is hyperthreaded, is ok even if it is only showing 50%. It is in fact using full capacity.
You might think about disabling the hyperthreading. Perhaps it's faster without it - on the other side if you only use half of the 'cores' it might doesn't matter.
joblack is offline   Reply With Quote
Old 2010-03-01, 04:04   #8
SO7783
 
Nov 2003
77377/77477 (either way)

22·3 Posts
Default

Quote:
Originally Posted by sdbardwick View Post
They are not using the same resources. C2D is using 100% of one physical processor, i5 is using 100% of two physical processors. Hyperthreading presents in Windows as two additional processors, so it reports 50% [(100+100+0+0)/4] overall utilization.
You are still switching things and not seeing what I've written both times: one out of 2 is fifty-percent, and two out of 4 is also fifty-percent.

And every other test, every other program, every other process treats these two machines identically (even Prime95 for all the other tests except this special instance)

I want to test 1 number. I want it to run at 50%. And I've done it for years. And years. And years. And I fully know how to play with affinity and running it as a service and/or multiple instances if that's what I wanted to do.

But all my other foreground and background programs handle things very well at 50% rather than 100% and I guess I'm never going to get a satisfactory answer on how the i5 is able to pull low wattage, run cool and quiet, and burn through the same task in half the time as an otherwise comparable core 2 duo.

SO7783 is offline   Reply With Quote
Old 2010-03-01, 04:12   #9
SO7783
 
Nov 2003
77377/77477 (either way)

22·3 Posts
Default

Quote:
Originally Posted by joblack View Post
You might think about disabling the hyperthreading. Perhaps it's faster without it - on the other side if you only use half of the 'cores' it might doesn't matter.
I've always felt the same way, although I'll always measure it with hyperthreading benefits/losses on a case-by-case basis to decide.

And this time, the i5 with 2 physical cores is getting twice the work done per unit time as the 2 physical cores Intel Core2(TM) is doing when both are at 50% testing one prime.

I have no doubt that disabling hyperthreading in this case would make the i5 650 machine take twice as long to complete as currently. This is the first hyperthreading free lunch I've ever seen -- I have always quickly disabled it for my purposes and OSes since the early 4 physical CPU XEON boards and the days of hot 3.06MHz P4s.
SO7783 is offline   Reply With Quote
Old 2010-03-01, 04:51   #10
petrw1
1976 Toyota Corona years forever!
 
petrw1's Avatar
 
"Wayne"
Nov 2006
Saskatchewan, Canada

455210 Posts
Default

Quote:
Originally Posted by axn View Post
Core i5, since it is hyperthreaded,
Point of correction: the i5 is NOT a hyperthreaded CPU...unless you mean you are using 2 threads for 1 Prime95 assignment....
petrw1 is offline   Reply With Quote
Old 2010-03-01, 05:05   #11
sdbardwick
 
sdbardwick's Avatar
 
Aug 2002
North San Diego County

2·11·31 Posts
Default

Quote:
Originally Posted by petrw1 View Post
Point of correction: the i5 is NOT a hyperthreaded CPU...unless you mean you are using 2 threads for 1 Prime95 assignment....
Some are
sdbardwick is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Optimizing Core2 quad in Windows XP John Rheinstein Hardware 18 2009-09-23 16:14
L2 cache size NOT RECOGNIZED for Core2 E8400 spartanroc Software 7 2008-10-03 15:34
Intel core2 Duo sieving? cipher Twin Prime Search 15 2007-06-05 21:20
Another Core2 Duo question Ender Hardware 3 2007-02-08 00:12
Core2 X6800 Test Times PrimeCrazzy Hardware 9 2006-08-29 08:34

All times are UTC. The time now is 10:09.

Sat Feb 27 10:09:11 UTC 2021 up 86 days, 6:20, 0 users, load averages: 1.72, 1.63, 1.55

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.