Still on AM = 3392 for my 6-core 5820K.

 

@Steve :

3392 is working fine in v2. Anything need to be changed for v3 ?

Should continue the same Gerard.  :smile:

I am not sure i see any different with HT=ON and AM=254 . Have you tried HT=off ? i seemed to have at least same FPS

That would be correct, only systemwide multitasking improves with HT enabled.

 

...with HT=Off the first sim thread runs on the first core (0), the next thread runs on the next core (1), the third thread runs on the third core (2), nice...

 

But with HT enabled we get better overall system efficiency.

 

However, look at how the sim starts with HT=On; the first sim thread runs on the first Logical Processor (LP0 core0), the next thread runs on the next Logical Processor (LP1 sharing core0), the third thread runs on the third Logical Processor (LP2 core1).

 

Each core ultimately will be multitasking with other threads, but it helps if we don't stack secondary sim threads onto the same core running the primary thread. So we use an AM (254 in your case with 8LPs) that masks one of the first LPs and the second thread moves onto the second core.

 

If we all use and understand the proper nomenclature; "cores" on an HT enabled "CPU" are presented as virtual processors called "Logical Processors", there would be less confusion. HT is done so that there is no impact when one core shares (time slices) two threads, so that stopping storing retrieving work done is halved during time slicing processes.

That would be correct, only systemwide multitasking improves with HT enabled.

 

...with HT=Off the first sim thread runs on the first core (0), the next thread runs on the next core (1), the third thread runs on the third core (2), nice...

 

But with HT enabled we get better overall system efficiency.

 

However, look at how the sim starts with HT=On; the first sim thread runs on the first Logical Processor (LP0 core0), the next thread runs on the next Logical Processor (LP1 sharing core0), the third thread runs on the third Logical Processor (LP2 core1).

 

Each core ultimately will be multitasking with other threads, but it helps if we don't stack secondary sim threads onto the same core running the primary thread. So we use an AM (254 in your case with 8LPs) that masks one of the first LPs and the second thread moves onto the second core.

 

If we all use and understand the proper nomenclature; "cores" on an HT enabled "CPU" are presented as virtual processors called "Logical Processors", there would be less confusion. HT is done so that there is no impact when one core shares (time slices) two threads, so that stopping storing retrieving work done is halved during time slicing processes.

 

Thanks , make sense

 

Its just i have seen NYxxUK having a relative huge FPS boost using HT and i dont see anything at all . Might be something else on his system.

 

So i will just leave with HT=off since i dont use much else during flight on my PC (ASN on a network PC) and can boost to @4,7 or 4,6GHZ

 

Michael Moe 

Disk I/o and networking are highly multi threaded, HT enabled contributes to a more efficient platform from which to run the sim. Some thread blocking and waiting problems can arise on complicated setups with few processors which are alleviated with HT enabled doubling the processor count. Take an extreme example with the dual core CPU. Results of HT On or Off are highly system specific.

Here is a question: I tried HT=on and AF=254 which leaves Core0 free. And HT=off and AF=14 which leaves Core0 free.

 

I found the FPS marginally better on maxed out setting (I should have used low setting in the test so that the difference between the two would have been more noticeable). However, I found that in both cases that Core1 was hovering around 98, 99%

 

My question then is why does HT=on not cause a performance bottleneck sooner than HT=off and yield lower performance?

I tried HT=on and AF=254 which leaves Core0 free. And HT=off and AF=14 which leaves Core0 free.

HT=On and AM=254=(11,11,11,10) leaves LP0 free, but LP1 (core0) is occupied by the main sim thread, so core0 is not free. With HT=Off and AM=14=(1110) that does leave core zero free.

 

To leave core0 free with HT=On, use AM=248=(11,11,10,00), leaves LPs 0 & 1 (core0), and LP2 (core1) free - thus leaving core zero free and masking one LP of the second core (core1).

Thanks for the answer :-) I see that with HT=on I should be using the term Logical Processor rather than Core. With HT=off I suppose that Core0

and LP1 are much the same thing. So I would have to surmise with HT=on despite Core0 (or perhaps better terminology would be physical processor0) LP1 still uses the full processing power of PP1?

more or less...except that more threads are spawned with more processors. They might gather information faster and better utilise the core. When we check the graphs we often see one processor maxed and the others with 20% occupied. On the other hand, with too many there'll be too much syncing with the main thread, so we can also mask processors we don't need - we might devote 6 cores from an 8 core CPU.

I guess the question that comes to my mind Steve is - how does this relate to the way that P3D uses the processor? On a 3770 at 4.7 - is 248HT on or 254HT on better and why? Or does it relate more to individual settings in sim?

 

I understand what you are saying about LP syncing etc but how does it relate specifically to P3D.

 

Personally I've been using 254HT on with good results - just switched to HT off no AM and trying that for a while to see if I notice any appreciable difference.

 

Vic

With more cores, and subsequently LPs, we can't improve the full on maxed speed of the one main thread. More LPs being available are good for the whole process anyway since upward of 40 threads are spawned within the affinity of the sim.

Hey Steve - 

 

I'm running a 4790K at 4.6 with HT off. I just set my AM to 12 so P3D uses only the last two cores. When I check the task manager for cpu load, it looks like core 2 is maxed out, and the last core is about 30% or so. I also tried an AM setting of 14, but core 1 was maxed out and the others would only go to about 30% as well. Why isn't the load being evenly distributed? Granted, with this setting, I do get fairly smooth results. 

The sim is limited by the speed of the main thread most of the time, the other threads gather information for the first thread and sometimes they get busy and at other times they slacken off.

I've just spent about two hours tweaking V3, as I did have some success tweaking V2. I've tried setting an affinity mask, as well as experimenting with FFTF settings.

In V2, the best results on my machine were achieved with no affinity mask, FFTF = 0.01, and a locked framerate. In V3, the same settings did result in a noticeably higher framerate (20-25% increase), although texture loading was somewhat slowed down. BUT: the overall perceived smoothness didn't improve, in fact it was slightly worse with the tweaks applied.

With the default settings in the CFG, I'm seeing smoother overall performance, despite what the fps counter says. It seems the people at LM did a good job in optimizing performance... It's almost as if professional software developers have a better idea of how to set up the sim than the average user, such as myself!

 

The only tweak I now have in my CFG is the most important one - the ultimate tweak, which not only increases performance beyond measure, but also improves the flying experience in general...:

FrameRate=0,0