When testing the sim we need to look at it in several ways. We have the ultimate fps capability, we have the speed in which it loads the scenario, and we want to know how smoothly overall it runs. We can use a stopwatch to see how the effect of an AM or other setting changes the loading times. Good loading times should give smoother sim operation in general. We might find that in the end we lose one or two fps but have the smoothest ride so ultimate fps in itself is not actually what we want, because all the data gathering processes have to stop the sim (generally referred to as Synchronize) to communicate their data. If we have too high an fps it might not have time to load the data fast enough results in momentary black textures or something like that.

We can turn off all the fireworks and set the texture size small and turn off the AI traffic, maybe turn off the weather too, and test it like that set unlimited and no vsync or limits, we can record the fps just by starting a flight at our fave airport right on the runway and leave it there, don't look around or do anything, or move any controls. We should get spectacular fps the CPU flying along not waiting for anything, and fairly steady too. Then we can increase for example the texture sizes and this if you can see it will start to test out a different part of the system, storage speed related, when before we had less effect from that. We find the same AMs tested don't show such big changes to fps because we now have inserted another limit to the performance. When we turn up all the GPU functions we go again and see less changes no matter what CPU arrangements because we red-lined it in another area. So testing out this stuff for more exacting results requires a lot of effort in small chunks. None the less Robs recordings of the entire flight for comparison are equally as useful although it doesn't divulge the details to readily it was a good test using the same setup and flight file. Once we start flying don't need to accelerate much differently each test to affect the results of what looks like faster and slower loading.

If we red-line it in some ways we actually can get a kind of smoothing out of the sim and this might look good but in fact we are deceived and could find a better setup in those cases.

 

Just started with a fresh install of P3Dv4.2

The 7940x I gave an Am of 2046 ( 00011111111110 ) , so 10 cores like Hasse , but without HT on as imho 20 Lp's would be too much.

 

At this moment I have settled for the following setings :

Core 0 = W10 = 4.7

Core 1 = Main thread = 4.7

Core 2 = Fiber thread = 4.4

Core 3-10 = Terrain = 4.3

Core 11-13 = Addons = 4.1

 

I need a good Core temp utility, but one that shows all 14 cores. The utilities that I have found are up to 6...

Any suggestions ?

 

 

 

 

Edited March 16, 20188 yr by GSalden

try coretemp http://www.alcpu.com/CoreTemp/

I think I mentioned this about HT once...HT on or off you have the same amount of cores - exactly the same and you  can count the threads on them the same - only difference with HT off you don't halve the switching time of threads with HT on you save on switching threads. I would have HT on and apply the appropriate AMs and ensure I don't have double process counts. With lots of cores HT on or off you still get processes using up too many. HT on increases heat because it does more work per cycle.

Think simplify: If you look at two threads on a non HT core and then try the same two threads on an HT core they complete earlier in HT mode. You got those same two threads irrespective of HT on or off.

Edited March 16, 20188 yr by SteveW

So if you got trouble with a 6 core with HT enabled you'll have the same trouble with 12 cores HT disabled.

16 minutes ago, SteveW said:

I think I mentioned this about HT once...HT on or off you have the same amount of cores - exactly the same and you  can count the threads on them the same - only difference with HT off you don't halve the switching time of threads with HT on you save on switching threads. I would have HT on and apply the appropriate AMs and ensure I don't have double process counts. With lots of cores HT on or off you still get processes using up too many. HT on increases heat because it does more work per cycle.

Won’t P3D get confused with 20 Lp’s with HT On compared to 10 with HT off ?

And the amount of extra heat with 28 Lp’s total.

No.

P3D does not get "confused". 10 cores HT off is 10 cores doing less work than they can with HT on and taking more time switching over the threads running on them - you can count the same numbers on or off. It's our job to apportion them to appropriate LPs. If you want to push the overclock frequencies then you can turn off HT to do that it slows down switching of threads. So you have to work out the compromise that suits you best.

 

Turning off HT "to save using too many cores" does not save cores at all and makes no sense to a CPU engineer. Only your AM can keep your apps on less cores.

 

 

 

Edited March 16, 20188 yr by SteveW

So AM = 4194300 .. ?

0000001111111111111111111100

Edited March 16, 20188 yr by GSalden

1 minute ago, GSalden said:

So AM = 4194300 .. ?

0000001111111111111111111100

That would be a terrible choice.

With P3D and FSX they make special processes that they keep to particular LP we show them with the mask we put in the cfg [JOBSCHEDULER] section..

P3D makes over 56 threads and they draw on a system that is probably running 200+ threads - all spread over your 14 cores. If they are spread over 28 LPs they are still spread over 14 cores the same number of threads and use the same amount of L3 cache as some like to go on about L3 a lot.

 Remember also that No AM is still an AM just enables all cores or LPs.

So those special processes we keep to a single LP of a HT core 01 instead of 1. So if anything else starts on that HT core it will start on 10 and save switching time. If it starts on the non HT core 1, it does not save the time.

So we apply masks to apps especially now as we got more than six cores.

 

Trouble is understanding apps that work like FSX and P3D is a little harder than usual.

 

Take one type of app might want as many LPs as it can get e.g. video rendering, it would probably spawn 14 processes all in parallel on that CPU. it would also (unattended) spawn 28 on that CPU with HT enabled.

 

With P3D on the one hand we want to save on that switching time, but on the other we don't want P3D (or FSX) to split out some processes only to put them back on the same core. So we utilise the AM that shows 01, for the monolithic tasks rather than 11 for parallel tasks.

 

So assuming you got no addons or apps and background tasks that hoover up all the available LPs willy-nilly you should be better off with an AM containing this in the middle '01,01,01,01,01,01', and set the overclock to accommodate the heat generated from the extra work done - unless you are pushing for a big overclock frequency then leave HT off.

 

22 minutes ago, GSalden said:

So AM = 4194300 .. ?

0000001111111111111111111100

Where you put all those ones and zero's shows no commas and looks like a non-HT mask string - better to put in the commas when we talk about HT enabled:

00,00,00,11,11,11,11,11,11,11,11,11,11,00

so now we can make sense of that arbitrary string of ones and zeros is in fact pertaining to HT enabled. Looking at all the double ones shows the mask has been chosen in error for an app like P3D or FSX and the rightmost cores should look like '01'.

 

With FSX we could gang up the back end processes on our four cores HT enabled with AMs like 11,11,11,01 - see on the left we have double ones because those will have FSX (and P3Dv3) processes that can run together on the same core.

The rightmost one of that string is the main sim thread, P3Dv4 is the same on the right end of the string - but P3Dv4 is different on the left end!

And so you guys need to study P3Dv4 a bit longer than we did FSX, but it's not too hard.