Hyperthreading


log in

Advanced search

Message boards : Number crunching : Hyperthreading

Author Message
Michael Agius
Send message
Joined: 22 Jan 00
Posts: 1
Credit: 6,842,637
RAC: 0
Australia
Message 1297344 - Posted: 20 Oct 2012, 21:58:15 UTC

Can anyone tell me from experience if hyper-threading has any advantage to the number of wu's processed over a given period of time? Is it worth the extra expense of having an i7 cpu instead of an i5 cpu from the same family of processors running at the same clock speed and running at 100% usage?
____________

Grant (SSSF)
Send message
Joined: 19 Aug 99
Posts: 5697
Credit: 56,427,096
RAC: 48,999
Australia
Message 1297347 - Posted: 20 Oct 2012, 22:02:08 UTC - in response to Message 1297344.

Can anyone tell me from experience if hyper-threading has any advantage to the number of wu's processed over a given period of time?

Yep.


Is it worth the extra expense of having an i7 cpu instead of an i5 cpu from the same family of processors running at the same clock speed and running at 100% usage?

Something only you can decide, but it's generally good for a 30-50% improvement, depending on what else the system is doing.
What really boosts computing performance is using a supported GPU.
____________
Grant
Darwin NT.

Profile James Sotherden
Avatar
Send message
Joined: 16 May 99
Posts: 8549
Credit: 31,435,525
RAC: 57,491
United States
Message 1297353 - Posted: 20 Oct 2012, 22:26:28 UTC

Im running an I7 920 with hyperthreading. Any time lost crunching a work unit is made up by running 8 at a time instead of 4. Plus using lunatyics more than makes up that difference. And with my GTS 250 I should be at over 10,000 rac just onthis machine alone. But Seti goes down everytime before I can get a good read on what this thing can do.
____________

Old James

Profile Tron
Send message
Joined: 16 Aug 09
Posts: 180
Credit: 2,236,055
RAC: 0
United States
Message 1297434 - Posted: 21 Oct 2012, 5:28:41 UTC

Here is something interesting I've noticed.

On my I7 860 2.8 ... within the preferences of boinc ,if I change the % of cpus to 75% and run cpu benchmarks,
the mips per cpu is significantly greater than if I set it at 100%

Greater still if I set an even lower percent.

Grant (SSSF)
Send message
Joined: 19 Aug 99
Posts: 5697
Credit: 56,427,096
RAC: 48,999
Australia
Message 1297486 - Posted: 21 Oct 2012, 8:27:51 UTC - in response to Message 1297434.

Here is something interesting I've noticed.

On my I7 860 2.8 ... within the preferences of boinc ,if I change the % of cpus to 75% and run cpu benchmarks,
the mips per cpu is significantly greater than if I set it at 100%

Greater still if I set an even lower percent.

And the Average Processing Rate for my GPUs is lower the more WUs that run on them.
But the fact is that running 3 WUs at a time on my GTX 560Ti does more work per hour than running 2, even though when running 2 the APR number is higher.
Same for the GTX 460- running 2 at time gives a lower APR value, even though it processes more work per hour than when just running one & the APR value is higher.
____________
Grant
Darwin NT.

msattler
Volunteer tester
Avatar
Send message
Joined: 9 Jul 00
Posts: 38327
Credit: 560,734,112
RAC: 655,548
United States
Message 1297491 - Posted: 21 Oct 2012, 8:41:29 UTC

Only my experience, YMMV always applies.
But, I think hyperthreading does not work up to it's hype.
I did extensive testing on a real high strung rig a couple of years back, and it did NOT pan out. Not much lost, mind you, but nothing gained at ALL. The rigs was much happier, more responsive, and I think turned in a bit more work running the tasks at full speed rather than half speed.
Which is about all one accomplishes by hyperthreading.

Just the kittyman's humble opinion and experience.

Like I said, YMMV.......the only way to know for sure on a given rig is to test it live. But you have to monitor the results very carefully.

If you don't wanna take the time and trouble to do that, take my advice and turn HT off.
____________
*********************************************
Embrace your inner kitty...ya know ya wanna!

I have met a few friends in my life.
Most were cats.

Profile Chris S
Volunteer tester
Avatar
Send message
Joined: 19 Nov 00
Posts: 31146
Credit: 11,325,566
RAC: 20,463
United Kingdom
Message 1297638 - Posted: 21 Oct 2012, 16:30:56 UTC

On a high end P4, turning on HT gives about a 10% increase in output. I cannot comment on the Intel i series processors.

Profile Mike
Volunteer tester
Avatar
Send message
Joined: 17 Feb 01
Posts: 23389
Credit: 31,832,232
RAC: 24,050
Germany
Message 1297646 - Posted: 21 Oct 2012, 16:52:41 UTC
Last modified: 21 Oct 2012, 16:53:12 UTC

I totally agree with Mark.
If you run GPU also its better to switch HT off.
____________

Profile Tron
Send message
Joined: 16 Aug 09
Posts: 180
Credit: 2,236,055
RAC: 0
United States
Message 1297651 - Posted: 21 Oct 2012, 17:17:18 UTC - in response to Message 1297486.

Here is something interesting I've noticed.

On my I7 860 2.8 ... within the preferences of boinc ,if I change the % of cpus to 75% and run cpu benchmarks,
the mips per cpu is significantly greater than if I set it at 100%

Greater still if I set an even lower percent.

And the Average Processing Rate for my GPUs is lower the more WUs that run on them.
But the fact is that running 3 WUs at a time on my GTX 560Ti does more work per hour than running 2, even though when running 2 the APR number is higher.
Same for the GTX 460- running 2 at time gives a lower APR value, even though it processes more work per hour than when just running one & the APR value is higher.


I was Referring to CPU benchmarking in my post, not talking about gpus just to be clear .. What I was pointing out was that intel has it's own "load control" at play here and the boinc preferances are really more of a sugestion to the host than actual hard policy.

I agree with Mark and many others that say hyperthreading is a bunch of hype.
My own experiments show hyperthreading gave a very slight gain in work production on some tasks while taking longer on others.
Overall the machines ran smoother ,and more stable at higher clock rates than with HT enabled.
I could do substantially more work with HT disabled and the MHZ cranked up, than I could do HT enabled and a lower (close to stock) clock rate

Wembley
Volunteer tester
Avatar
Send message
Joined: 16 Sep 09
Posts: 415
Credit: 888,257
RAC: 0
United States
Message 1297768 - Posted: 22 Oct 2012, 0:38:33 UTC

If you only have one core, then turning HT on gives a slight boost to overall crunching performance, but slows down other things that occasionally need 100% of the CPU since it will only get half a CPU max. If you have more than one core, HT is useless.
____________


Donate with your searches and online buys:
http://www.goodsearch.com/toolbar/university-of-california-setihome

Profile HAL9000
Volunteer tester
Avatar
Send message
Joined: 11 Sep 99
Posts: 3865
Credit: 107,125,257
RAC: 99,069
United States
Message 1297952 - Posted: 22 Oct 2012, 19:09:47 UTC

In testing the 6 machines I have with i7-860's I tried various configurations.
HT Off: 4 tasks at once.
HT On: 8 tasks at once, & 4 tasks at once.
HT On with 8 tasks was the top producer, but HT On and 4 tasks was only a few % behind it. HT Off with 4 tasks was several % behind HT On with 4 tasks.

The 2nd & 3rd generations chips may produce different results.
____________
SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

Message boards : Number crunching : Hyperthreading

Copyright © 2014 University of California