Mike, your monitor is far far beyond my depth. But what I noticed on my 31' NEC running at any refresh I want, that soon as I limit FPS in P3D to anything from very modest to very high numbers I start seeing micro micro stutters of the terrain all the same, particularly in the turns and in congested areas and heavy wx, of course. But at VSYNC ON, FPS Unlimited, low monitor refreshes 20, 25, 30 - my system runs Prepar3D v4 so smooth! Now I don't know why I bought my binned @4.7GHz and delidded i9 7900x, I should have started learning HT vagaries and benefits a little earlier... 

Thanks!

 

Edited January 19, 20188 yr by Dirk98

5 hours ago, SteveW said:

a four core HT ON 01,01,01,01=85 does the same job four full on cores

85 was not the best AM I tried on my 6700k. Now I think that giving P3Dv4.1 just 4 LPs is not enough! Add-ons or no add-ons.

Thanks.

With FSX and 32bit P3D four cores was hard to beat, but v4 can and will use more, 6 or even 7 are effective, the way the parts split out then is not so intuitive. Remember though that using more LPs for one thing is a bigger slice of the whole CPU resource leaving less for others to get other work done..

Quote from 2 days ago:

"Just finished my flight from Amsterdam Schiphol to Milan Linate. Heavy weather. Just after take of and after the last turn to approach I saw a short stutter.  

Today was with HT on and the addons spread over LP cores : 3+5 , 7+9 , 3+7 and  5+9."   ---> AM=1365  / CPU all cores at 4.3 Ghz

As you can see the terrain coes are nor maxed out. Therefore I have spread my addons over them and they still are not taxed that much. 

 

 

This morning I did the same flight again with HT off and no AM. Addons spread over the last 5 cores ( each addon over 2 cores )

I used Historical weather form ASP4 with the same time to get as close to the former test flight.

I experienced the same stutter just after take off , no after the last turn to approach , but several small jitters at approach .

---> CPU core 0 at 4.4 GHz core 2-5 at 4.3 GHz

 

With both I had a good flight. There were big cloud structures and Milan was almost covered in fog ( 300 ft and upwards )

 

Steve, any suggestions ?

  

  

 

4 minutes ago, GSalden said:

Quote from 2 days ago:

"Just finished my flight from Amsterdam Schiphol to Milan Linate. Heavy weather. Just after take of and after the last turn to approach I saw a short stutter.  

Today was with HT on and the addons spread over LP cores : 3+5 , 7+9 , 3+7 and  5+9."   ---> AM=1365  / CPU all cores at 4.3 Ghz

....

Steve, any suggestions ?

  

  

 

I'm not Steve by any means, but I would try 1366 as well!!  ))

Thanks!

Edited January 19, 20188 yr by Dirk98

6 minutes ago, Dirk98 said:

I'm not Steve by any means, but I would try 1366 as well!!  ))

Thanks!

1366 = 01 01 01 01 01 11

Why would I benefit from this over 1365 ?  If core 0 is full it will use core 1 as well.

22 minutes ago, GSalden said:

1366 = 01 01 01 01 01 11

Why would I benefit from this over 1365 ?  If core 0 is full it will use core 1 as well.

1366 = 01 01 01 01 01 10

Some posts earlier I noticed Steve often suggested 10 on the first Core. But without much going into details why 10 not 01 ))

Edited January 19, 20188 yr by Dirk98

1 hour ago, Dirk98 said:

1366 = 01 01 01 01 01 10

Some posts earlier I noticed Steve often suggested 10 on the first Core. But without much going into details why 10 not 01 ))

Surely Steve can explain why and if 1366 might be better than 1365.

Hi Gerard, While it means nothing to P3D "01" or "10" what's going to happen within the system is not entirely certain because it looks at core usage statistically moves from LP to LP to find less activity to distribute work to. So it's easiest to think of it in a simple way. Say with P3D on 10,10 and you are looking at LP1 (second from right) probably busy with a sim process - move on. The next LP is the next actual physical core. Starting with 01,01 and you are looking at LP1, so what if it's a zero it could be very busy with system stuff, move on, LP2 (third in from right) a sim allocation might not look so busy at the time and the process goes there. So although it makes no difference to P3D 01 or 10, the way things pan out can change. It's stuff like that means I'm never able to say explicitly use ABC and 123.

Thanks Steve , for explaining.

Looking at my HT on pic above , to me it looks nicely devided over the cores.

Surely I know that with AM 1365 Windows and P3D main thread are on core 0, but it is running very nicely ( triple view + framerate locked inside at 20 ).

I will try again with HT on and AM 1365 , but will boost core 0/1 to 4.4 GHz and leave the other at 4.3 Ghz.

Then I can compare better with no HT and core 0 at 4.4 Ghz + core 1-5 at 4.3 Ghz.

2 minutes ago, GSalden said:

Thanks Steve , for explaining.

Looking at my HT on pic above , to me it looks nicely devided over the cores.

Surely I know that with AM 1365 Windows and P3D main thread are on core 0, but it is running very nicely ( triple view + framerate locked inside at 20 ).

I will try again with HT on and AM 1365 , but will boost core 0/1 to 4.4 GHz and leave the other at 4.3 Ghz.

Then I can compare better with no HT and core 0 at 4.4 Ghz + core 1-5 at 4.3 Ghz.

Just to make sure, you're not going to test 1366, right? Or you misspelled it by 1365?

40 minutes ago, SteveW said:

Hi Gerard, While it means nothing to P3D "01" or "10" what's going to happen within the system is not entirely certain because it looks at core usage statistically moves from LP to LP to find less activity to distribute work to. So it's easiest to think of it in a simple way. Say with P3D on 10,10 and you are looking at LP1 (second from right) probably busy with a sim process - move on. The next LP is the next actual physical core. Starting with 01,01 and you are looking at LP1, so what if it's a zero it could be very busy with system stuff, move on, LP2 (third in from right) a sim allocation might not look so busy at the time and the process goes there. So although it makes no difference to P3D 01 or 10, the way things pan out can change.

!!

Thanks.

Waste of time to assume WIndows is on LP0 or core zero. All cores are equal, some core Windows stuff may start up on zero but it stops and restarts so if it's finding it's place busy from an app the thing moves to another slot.

So it must be pretty obvious now that if the sim splits out a process to utilise another core then if we have HT enabled and no AM we get 11,11 etc. And so the split out moves to the next LP of the same core and not the intended next physical core. It's OK for the split out process to run on the other LP of the core, it's better than the one thing on one non-HT core that's for sure. So we do 01,01 (or some variation one LP enabled per core) and the split-out is guaranteed to move to another physical core, and not simply being the sister LP of the same core.

With many cores why gang up stuff doubling sim work per core with sim AMs containing 11,11 like no AM? We wouldn't normally if we had 8 or more cores we can partition it. With less cores we can put some parts of the split out two per core 11,11 if it's not the main process -remember that's the sacred one.

With many cores we find the optimum amount for the sim, and then partition it off leaving the rest for  targeting addon exe apps.

Slinging more and more cores at the sim will reach a point our hardware can't handle, it will continue to split out jobs and each starts waiting for the other - keep it lean.

 

 

 

10 minutes ago, SteveW said:

Waste of time to assume WIndows is on LP0 or core zero. All cores are equal, some core Windows stuff may start up on zero but it stops and restarts so if it's finding it's place busy from an app the thing moves to another slot.

Then I will now use 1365 with Core 0/1 at 4.4 GHz and the others at 4.3 GHz. My system has liquid cooling so heat will be of no issue.