Skynet HT question & my new i7 rig....

[Assimilator1]

Finally got around to building my i7 rig on Sunday (after having to wait for a new adapter kit to be released for my HSF amongst other things), the bent pin (which I more or less straightened) doesn't seem to be causing a problem.... so far!

Anyway I've got an Ivybridge S2011 i7 4820k which has HT, it's at default clock of 3.7 GHz (until I confirm the dodgy pin isn't causing any problems).

ATM I'm running Skynet POGS with HT on & I'm getting times of ~11,200s for 303 credit WUs.
My old C2Q @3.6 GHz was getting ~11,000s (+/- 100s) for the same 303 credit WUs but of course it wasn't doing 7 at once (1 core saved for GPU crunching).

Anyway, anyone compared HT on & off for Skynet?

 

[TennesseeTony]

I don't have any recent tasks to compare times with, but that sounds about right, for HT enabled. I've not run it without HT.

I doubt the pin is going to be a problem, as it currently is working.

 

[Assimilator1]

Ah ok cool, I am curious as to what times I'd get without HT though, is the only way to disable it via the BIOS?
Or will setting BOINC to 50% of the CPU achieve the same?

 

[TennesseeTony]

BIOS

 

[Assimilator1]

Ok thx mate , I wonder if I can access that part of the BIOS in windows, I know they're some things I can.....

No one else got thoughts/figures for Skynet & HT?

 

[waffleironhead]

I havnt run any skynet. Have you tried just reducing available cores in boinc to 50% and seeing what the WU times are?

Is it really necessary to turn of HT in bios?

 

[Assimilator1]

I could but I don't know if the tasks would be running on real or virtual cores.

Know any programs that'll show me that?

 

[waffleironhead]

In my experience window's thread scheduler seems to know the difference between physical and logical? cores. I'm just not sure if there is any performance issue with leaving the ht cores on. Though, even if the particular program does not benefit from ht, the ht cores are fine for feeding gpus so maybe you would be better off leaving them on.

 

[Assimilator1]

True, but like I said I'm just curious as to what times I'd get without HT , would be nice to know the actual difference in output with HT on in Skynet.

I don't see window's thread scheduler in TM, where is it?

 

[Assimilator1]

Google didn't seem to help.....

 

[TheELF]

In my experience window's thread scheduler seems to know the difference between physical and logical?

No the HT cores are transparent to every windows program,they only see them as cores,you need some very speciallized intel developer software to see whats running where.
Thing is you can assign affinity for the programm from task manager and let it run on only 4 cores,since HT only uses commands left over by the thread that is running on that core it will be no difference if it runs on real or HT since if theoreticaly the thread runs on HT then 0% of the real core would be used so 100% of HT core would be used.

I don't know how confusing that sounded but basically if the real core doesn't do anything then HT has 100% of resources available.


Good read on HT,also tells you ways to distinguish between Cores and Intel HT Technology Threads
https://software.intel.com/en-us/articles/performance-insights-to-intel-hyper-threading-technology

 

[Assimilator1]

Ok, thanks for the info .

I'll just disable HT to test WU times without it then.

 

[waffleironhead]

Yes, but my understanding is that windows reads data from the cpuid, and assigns programs to the cores based on if HT is there. If HT is available windows will load core 0,2,4,6 before loading core 1,3,5,7. Its been that way since XP.

 

[waffleironhead]

I just did a few test runs on some same sized wu.(56433 GFLOPS)
Running boinc at 50% available processors: 90 minutes/WU
Running boinc at 100% available processors(HT cores working): 113 minutes/WU

Looks like HT is a benefit on Pogs.

 

[Assimilator1]

Cool, will be interesting too to see if their's any difference between BOINC at 50% & having HT off.