Author Topic: Calcimage faster with hyperthreading disabled  (Read 3475 times)

Philipp Poeml

  • Professor
  • ****
  • Posts: 219
Calcimage faster with hyperthreading disabled
« on: April 18, 2016, 02:56:52 AM »
I have something interesting here. I am testing my new i7 PfE PC and was doing some calculations. Unfortunately PfE is using only 1 core for everyting it does, so I figured, what happens if I switch of "hyperthreading", which is supposed to make computers faster? In the internet I learned that it should have no effect to switch it off because the modern hyperthreading is cool and is managing cores and threads very well.

I calculated the same calcimage project, once with hyperthreading on, once off. Result:
hyperthreading on: 19.5 minutes
hyperthreading off: 12.5 minutes

So there IS a difference. Amazing!

Now I am wondering if I should switch it off everywhere else too, because I don't really have applicatoins that use it. However on the laptop I have only 2 cores, i.e. 4 virtual ones with hyperthreading on. Since also Win7 64 bit is using some, and there I run more applications next to PfE I might leave it on there.

Will test.

John Donovan

  • Administrator
  • Emeritus
  • *****
  • Posts: 3280
  • Other duties as assigned...
    • Probe Software
Re: Calcimage faster with hyperthreading disabled
« Reply #1 on: April 18, 2016, 06:19:18 PM »
I have something interesting here. I am testing my new i7 PfE PC and was doing some calculations. Unfortunately PfE is using only 1 core for everyting it does, so I figured, what happens if I switch of "hyperthreading", which is supposed to make computers faster? In the internet I learned that it should have no effect to switch it off because the modern hyperthreading is cool and is managing cores and threads very well.

I calculated the same calcimage project, once with hyperthreading on, once off. Result:
hyperthreading on: 19.5 minutes
hyperthreading off: 12.5 minutes

So there IS a difference. Amazing!

Now I am wondering if I should switch it off everywhere else too, because I don't really have applicatoins that use it. However on the laptop I have only 2 cores, i.e. 4 virtual ones with hyperthreading on. Since also Win7 64 bit is using some, and there I run more applications next to PfE I might leave it on there.

Will test.

Hi Philipp,
Interesting.

If I were you I would perform a "back to back" test.  Because the 2nd test might be quicker just because it was run 2nd- for example because of disk caching).

So maybe try this:

1. Run with hyper threading
2. Run without hyper threading
3. Run with hyper threading
4. Run without hyper threading

Are 1 and 3 similar?  Are 2 and 4 similar?  OK. Now which set is faster and by how much? Then calculate the variance of each set...     :)
« Last Edit: April 18, 2016, 10:34:59 PM by John Donovan »
John J. Donovan, Pres. 
(541) 343-3400

"Not Absolutely Certain, Yet Reliable"

Philipp Poeml

  • Professor
  • ****
  • Posts: 219
Re: Calcimage faster with hyperthreading disabled
« Reply #2 on: April 19, 2016, 03:34:38 AM »
I could do multiple runs, yes.

However, on my computer the hyperthreading is switched off in the bios, which makes a restart necessary. So I don't think the 2nd test had any disk caching or anything. So before each test I made a fresh reboot of the system to have similar conditions.

John Donovan

  • Administrator
  • Emeritus
  • *****
  • Posts: 3280
  • Other duties as assigned...
    • Probe Software
Re: Calcimage faster with hyperthreading disabled
« Reply #3 on: April 19, 2016, 07:41:17 AM »
I could do multiple runs, yes.

However, on my computer the hyperthreading is switched off in the bios, which makes a restart necessary. So I don't think the 2nd test had any disk caching or anything. So before each test I made a fresh reboot of the system to have similar conditions.

If each image calculation was performed just after a re-boot, the test as a whole should be good.  Still might be worth trying both a small image and also a large image and see if the effect is size independent. 

I'd be interested if anyone has an explanation for the calculation being faster without hyper threading.
John J. Donovan, Pres. 
(541) 343-3400

"Not Absolutely Certain, Yet Reliable"