I'm going to test a scenario a few times this afternoon with different AM settings. For those who have decided on a particular AM setting did you set that objectively or subjectively?

I'm going to load a saved scenario with Concorde at EGLC, slew up to 1500ft and track westward towards Heathrow.

That eats the most VAS when I'm flying and fps are near their lowest so any differences in AM should hopefully indicate which is best. FSUIPC logs will show the average fps for the test which will help enormously.

I've conducted the same test several times. The only difference between tests is the HT status and AF setting.

I placed FS Labs Concorde-X at Rwy 28 at EGLC. Switched to 2D panel and hit W to show only 5 small instrument gauges at the bottom. I slewed up to 1500ft and turned to a heading around 250-260 degrees. Heathrow is just visible on the horizon.

I hit the forward key several times until 240kts was achieved. It was then just a case of minor adjustments to my heading so I passed over the Control Tower at Heathrow and the southern end of T5. As I passed the final reservoir west of the airport I hit 5 to stop slewing and exited P3D. All data below is taken from the FSUIPC log. Surprisingly similar results:-

Green indicates best result for that test; red, the worst. Even with the same test and conducted the same way there are variations as the 84 test proves.

There is a case for turning HT off! It also proves that staying with 255 was not a good move. I'm currently on 84.

Attempting to assess performance based on fps is a problem, because also we have background throughput. With not enough background throughput the sim can show higher fps because there's not enough happening in the data loading department.

VAS and background processing is consumed by adding in bad weather and AI traffic and other objects. My tests involve scenarios that stress the sim in different ways on a test setup that makes very repeatable tests with software designed specifically to find the best data. Remember the sim has to maintain fps irrespective of the background performance - anyone heard of the blurries?

If you've got a reasonable AM such as on the four core+HT AM=85 or three core AM=116 these provide the best performance in theory and practice - HT off won't touch those. However, messing around from those AMs won't produce better performance, instead a bigger impact is the arrangement of exe apps running alongside the sim. Don't be fooled by a little tiny exe making use of a little tiny 6% CPU time - it can still rubbish the sim.

Use a .bat to arrange the startup of other apps. Process lasso is OK but not efficient with many types of app when changing their affinity, especially apps like P3D and any that make their own Affinity arrangements.

5 minutes ago, SteveW said:

Attempting to assess performance based on fps is a problem, because also we have background throughput. With not enough background throughput the sim can show higher fps because there's not enough happening in the data loading department.

VAS and background processing is consumed by adding in bad weather and AI traffic and other objects. My tests involve scenarios that stress the sim in different ways on a test setup that makes very repeatable tests with software designed specifically to find the best data. Remember the sim has to maintain fps irrespective of the background performance - anyone heard of the blurries?

If you've got a reasonable AM such as on the four core+HT AM=85 or three core AM=116 these provide the best performance in theory and practice - HT off won't touch those. However, messing around from those AMs won't produce better performance, instead a bigger impact is the arrangement of exe apps running alongside the sim. Don't be fooled by a little tiny exe making use of a little tine 6% CPU time - it can still rubbish the sim.

Use a .bat to arrange the startup of other apps. Process lasso is OK but not efficient with many types of app when changing their affinity, especially apps like P3D and any that make their own Affinity arrangements.

Hi Steve, 

Thanks for the explanation. I'm currently running AM=116 on an i7 6700k with HT on. Would you say this is optimal? 

Thanks 

Hi, variants based on the 116 pattern for three core use is good because it utilises three cores more efficiently than any other setting. 85 or 170 (same pattern) will provide better performance since they use four cores - but not by much. With the three core variants it leaves a core free for addon exe apps. With four cores used we have to be more careful with the way these apps layout across the CPU, keep them on the last two cores whereby they share cores running background tasks only. In Rays examples he appears not to have tried 85 or 170 HT on for some reason but even so would not give any indication of true performance. Those results show a very obvious change in fps across the board which suggests problems associated with other processes are disrupting the flow of the sim and the AMs produce different results around that behaviour, that is all. They don't show throughput capacity since it's not linked directly to fps. 

...anyone in the HT Off department is losing performance specific to the way windows handles hundreds of threads. Don't believe the anti-HT hype, investigate why Intel/MS do it, why it costs more and how to gain from it.

 

21 minutes ago, SteveW said:

VAS and background processing is consumed by adding in bad weather and AI traffic and other objects. My tests involve scenarios that stress the sim in different ways on a test setup that makes very repeatable tests with software designed specifically to find the best data. Remember the sim has to maintain fps irrespective of the background performance - anyone heard of the blurries?

If you've got a reasonable AM such as on the four core+HT AM=85 or three core AM=116 these provide the best performance in theory and practice - HT off won't touch those. However, messing around from those AMs won't produce better performance, instead a bigger impact is the arrangement of exe apps running alongside the sim. Don't be fooled by a little tiny exe making use of a little tiny 6% CPU time - it can still rubbish the sim.

Use a .bat to arrange the startup of other apps. Process lasso is OK but not efficient with many types of app when changing their affinity, especially apps like P3D and any that make their own Affinity arrangements.

I didn't have many tools available to me to properly record stats. But better than a subjective test of course.

The only other processes running where the GoFlight module and EFB Dara Provider. I also ran with clear skies but usually use AS16/ASCA.

My main concern is not fps but VAS. All I can really do now is stick with 84 for a few flights and see how I get on. Things should be better than with an AM of 255.

Steve, I did run an 85 test but not a 170. Is it best if every other core is running P3D rather than two adjacent? So 01010101 is better than 01100101?

Hi Ray,

 

VAS won't change enough to worry you with changing AMs or HT because the sim rendering does the same thing on one core or several. As we reduce cores with an AM arrangement we merely gang up those processes on less cores, and so they also go onto the core with the rendering stage making it slower.

Read the P3D docs, these clearly show when the sim is split over four parts it runs the renderer at its leanest. Reduce or add cores and it's less efficient, but with more cores will load the scenario faster - that is all.

Don't think of cores, that's where the misunderstandings emanate from, instead think of Logical Processors. On the current desktop gaming CPUs there are two per core with HT enabled, some CPUs have 4 or more per core, wouldn't want to turn HT off with those, then why when there are only two?  lol.

So think LPs not cores - each pair of LPs shares a compute core but each have their own hardware backing them up to collect and arrange the code before it's computed. Turning off HT leaves you with only half that capacity and so makes less heat since it's doing less work - overclockers arrange the OC down the heat curve to accommodate HT on. The caching (another mystery subject) is unaffected as long as the CPU runs the same number of threads HT on or off. That's why the AM is important and been there from the start.

All the best

Steve

 

 

Hi Steve. My mistake, I meant virtual processors, not cores. My i4770K has 4 cores but 8 virtual processors, yes?

I've just realised AM of 84 is wrong. It's only using 3 VPs, not 4. 84 is 1010100. It needs to be 01010101 which is 85. I did conduct a test with 85 but the results were slightly worse than 84. Probably the test is the issue, not the AM.

Thanks for the explanation of things. It helps a lot.

yes that's right Ray; if 85 doesn't show the best result then it's going down to the way things pan out across the chip unfavourably...on the differences between say 85=01,01,01,01 and 170=10,10,10,10, or 116=01,11,01,00 and 184=10,11,10,00 - there's none as far as the sim is concerned. However the final layout of processes across the CPU is altered. So what's important is to give the sim four parts and arrange those addons to avoid the rendering stage (first two parts).

...If we had a braking problem from 80mph but was not too bad from 70mph, we wouldn't accept a cure that simply restricted our speed to 70, we would want more attention paid to the servicing of the brakes. If HT on and 85 doesn't give the best result, then you should be asking "why not", rather than changing AMs. Could be in the end one or more of your addon exe apps needs to be on a core of their own or they just clog that core and nothing the system can do about it.

Give addons at least two LPs. Here's an overview of setups for a 4core+HT using three and four cores:

The four core versions:

01,01,01,01 = 85 dec - sim - best rendering
10,10,00,00 = A0 hex - apps (for batch file setting)

or

11,11,01,01 = 245 dec - sim - best loading
10,10,00,00 = A0 hex- apps


The three core versions:

01,11,01,00 = 116 dec - sim - best rendering
10,00,00,01 = 81 hex - apps or
10,00,00,10 = 42 hex - apps or
00,00,00,11 = 3 hex - apps

or

11,11,01,00 = 244 dec - sim - best loading
00,00,00,11 = 3 hex - apps
 

21 minutes ago, Ray Proudfoot said:

Hi Steve. ...Thanks for the explanation of things. It helps a lot.

You are welcome Ray! My inbox rules funnel mail into two main subjects: 'I now have better performance', and 'I now have worse performance'. The split of mail, from arguably the most friendly and intellectual group on the planet, currently is 100% in favour of better.

Steve. Clearly you know your subject.  I look forward to a proper flight with the new AM. Slewing between airports is not a real test but was the only way I could test things.

We are fortunate to have people like yourself in these forums. 

If you ask me, having Affinity Mask off actually provides quite good and well-balanced values for average FPS, minimum FPS, maximum FPS and VAS consumption according to your table.

I have a four-core CPU in my system (i5 4690K). One time, I experimented with the AF values by assigning the cores 2,3 and 4 to Prepar3D. The motive was that without AF, I saw core 1 at nearly 100% in the task manager, while the other cores were sitting almost idle. Setting AF=16 provided a measurable boost to the initial FPS on runway, and the cores 2,3 and 4 were tasked evenly. However, after further experimenting, I noticed that the FPS on approach were lower than without using AF and that the autogen objects appeared later in some instances. So I deleted the AF entry and stopped messing with it. While with AF off, the other cores are unstressed in the initial state where hardly any scenery is loaded, you can see all four cores active when Prepar3D loads intensive scenery during a flight. In that situation, if you "remove" a core from P3D, you reduce the overall computing power and lower your FPS (where only three cores are available in a situation where all four cores were loaded without using AF). The default state without AF is not for nothing.