fastgeek
OK, seems I can't post links.  It's over in the [H] DC forum for those who care to look.
 

You can post links just leave off the http:// part
 
https://hardforum.com/threads/primegrid-isaac-newtons-birthday-challenge-part-1-of-9-of-primegrid-challenge-series.1921077/#post-1042739470
 
Thanks

howdy2u2
Well the i7 4790k with HT off is pretty much 2:1. So I am leaving HT off
 
I just turned off HT on my other 2 rigs i7 875k and a i7 870, I bumped up the clocks on the 875, and dummy OC the 870 since it's locked. Time will tell I guess, basically drops 8 threads off my WU participation. If it decreases the times to a ratio like the 4790k it will be worth it, temp wise on any my CPUs it only drops a few degrees. (all on water) I shall add that, as long as I produce error free WUs I'm happy!!
I know my rigs are old school, but they are converts from Folding!!!

 Why turn it off? You should leave HT on the 875k and 870, unless you just don't want to, that's ok too.
 
I just found one of my i7-3930k had HT off, just turned it on and bumped up the hz
 

tanner2
I think we need punchy to change his name to speed racer

I needed more pie!  

bill1024
Why turn it off? You should leave HT on the 875k and 870, unless you just don't want to, that's ok too.
 
I just found one of my i7-3930k had HT off, just turned it on and bumped up the hz
 

Well it's kind of early on for results with the HT off with the 870 BUT
 
HT on = 10,609.23 Sec/ 176.82 mins/ 2.94hrs (avg of 6 WUs) = 8.14 WU/ Day
 
HT off = 7965.20 Sec/ 132.75 mins/ 2.21hrs (avg of 6 WUs) = 10.84 WU/Day
 
At this point it's a ~45 minute quicker completion time/ WU < The times for HT off are still in progress since they were partially completed with HT on.
I need to let it run longer to get better results. If I am incorrect I will turn HT back on and run it the rest of the way like that.
It's not that I want to turn it off, but if the 4970k can run  2:1 with HT off and if these come closer to that ratio and not hurt the team in doing so I think they are better off running with HT off. So far I am holding the position I am at..................
 
If you folks think I need to turn HT back on, I will. You da bosses I'm just here for the money!!! 

fastgeek

bcavnaugh
 
"Still doesn't explain what the heck is going on with using cc_conf to restrict BOINC to 50% of the CPU; aka the lazy way of "disabling" hyper-threading."
 
This is not really true, you will still be running 2 tasks Per Core.

I don't quite follow what you're saying.  Could also be that my previous experience with using the "50%" method via cc_config worked almost identically to turning off HT in the BIOS.  Did this as it was easier to just modify cc and re-read preferences vs. rebooting and setting it in the BIOS.  

Because with HT On you will run Two Task Per Core (Two Threads), with HT Off you will only run One Task Per Core (No Threads).
It will run Core 0 Thread 0 and Thread 1 before it will start on Core 1 Thread 0 and so on. Intel® Hyper-Threading Technology
 
This is my  Xeon(R) CPU E5-2680 v4
4,035.12 HT Off
8,053.68 HT On
 

wizanhi
So my tests are as follow.  Only thing running is BOINC.  Folding is off so GPU was idle during this time.
 
i7-4790k stock setting at 4.0ghz. Temps <70 degrees.
 
HT-off
WU / Runtime (s) / CPU time (s)
508368501 / 3115 / 2962
508363273 / 3189 / 3065
508363037 / 3140 / 2998
 
HT-On
WU / RT (s)/ CPU time (s)
508380899 / 4640 / 4457
508379274 / 4794 / 4527
508381399 / 5577 / 5253
508381614 / 5489/ 5171
 
so HT on seems more beneficial for me.  Regardless, not a big enough difference to be worth changing any specs if you have HT/no-HT cpus set for other things.  
 
 

Yep! you should keep HT On.

Current operating systems are HT-aware and will avoid scheduling on both logical processors of a single physical core until it's necessary.  Setting BOINC to use 50% of the processors (NOT 50% of the CPU time) will give you pretty close to the same as turning HT off.

howdy2u2

bill1024
Why turn it off? You should leave HT on the 875k and 870, unless you just don't want to, that's ok too.
 
I just found one of my i7-3930k had HT off, just turned it on and bumped up the hz
 

Well it's kind of early on for results with the HT off with the 870 BUT
 
HT on = 10,609.23 Sec/ 176.82 mins/ 2.94hrs (avg of 6 WUs) = 8.14 WU/ Day
 
HT off = 7965.20 Sec/ 132.75 mins/ 2.21hrs (avg of 6 WUs) = 10.84 WU/Day
 
At this point it's a ~45 minute quicker completion time/ WU < The times for HT off are still in progress since they were partially completed with HT on.
I need to let it run longer to get better results. If I am incorrect I will turn HT back on and run it the rest of the way like that.
It's not that I want to turn it off, but if the 4970k can run  2:1 with HT off and if these come closer to that ratio and not hurt the team in doing so I think they are better off running with HT off. So far I am holding the position I am at..................
 
If you folks think I need to turn HT back on, I will. You da bosses I'm just here for the money!!! 

wait a minute...there's money involved ?!??!

bcavnaugh
So then like this;
Core 0 Thread 0 then Core 1 Thread 0 then Core 2 Thread 0 then Core 3 Thread 0 then Core 0 Thread 1 Core 1 Thread 1 and so on?

It's a little more random than that, but overall the effect is the same.  In Windows at least you will see some cores that "stick" to one thread, while others swap back and forth.  Since the two threads in a core share L1 caches, there is almost no penalty to switch.  I have no idea why they don't just "stick", but that's what I see.

Punchy

bcavnaugh
So then like this;
Core 0 Thread 0 then Core 1 Thread 0 then Core 2 Thread 0 then Core 3 Thread 0 then Core 0 Thread 1 Core 1 Thread 1 and so on?

It's a little more random than that, but overall the effect is the same.  In Windows at least you will see some cores that "stick" to one thread, while others swap back and forth.  Since the two threads in a core share L1 caches, there is almost no penalty to switch.  I have no idea why they don't just "stick", but that's what I see.

Thanks I am testing this now, my goal is that the time to complete one Task with HT on will be the same as if HT was Off.
 
I have to wait until all the current tasks are done to get a good complete time in this test.
 

New Today Generalized Cullen Woodall Sieve  Ruby

bcavnaugh
New Today Generalized Cullen Woodall Sieve  Ruby

awesome bcavnaugh, congrats

Sorry Guy's, All the problems with my 48 core server, are now over, it is dead.
I was trying to catch bcavnaugh, might had done it by the end of the challenge.
 
But not, It is down and out for the count. Not happy...........
 
Good news is I have everything else running at 1000%, not that is not a typo..

Noon Pie Time

Oh no, not the 48-core system!   So sorry.
 
Bcav, your system has so many threads it's hard to see the scheduling behavior.  Not that too many threads is a problem... On my dual 8-core system, it's a little easier to see.  In task manager the two threads of a physical core are together in left-right pairs.
 
On my system, the highest numbered cores don't swap between threads too much, for example the CPU 1 Core 7 pair.  You can see how the two threads are taking turns, so one's peaks are the other's valleys, on CPU 0 Core 6, and even CPU 0 Core 0 where you have to look closer.  The more often they switch, the harder it is to see.
 
I'm not sure why your system is showing 56% utilization.  Are you folding at the same time?  Whatever that 6% extra load is might be making things a little harder for the scheduler.
 
 
[sorry, can't figure out how to show the image inline...]