Thanks Rob. I've had sufficient enthusiastic feedback suggests getting good results with the settings I generally recommend. You won't be letting anyone down. On the money as usual though, Rob. Since it was very good of you getting in a note about the situation with the GPU. Where we must consider carefully, AA, VSync, fps aimed for, all play a vital role.

2. What is ExpandedFSAffinity (EFSA) and why is the default IF10 Pro setting=True?

Advanced affinity space setup for P3D, FSX, and FSX-SE. Here clearly demonstrates Ideal Flight 10 Professional setting ExpandedFSAffinity on P3D v3.2 and also IF10 app startup vs starting exe's from exe.xml.

 

EFSA gaining as much as 2% performance enhancement in a very demanding and strictly repetitive test here with P3D v3.2 on a 6 core HT enabled:

 

 

 

The result at the top could not be improved on no matter what setup could be implemented. Utilising more than four cores for P3D dropped performance slightly. P3D v3.2 appears to perform slightly better with problem CPU set-ups than v3.1.

 

Notice how the sim takes around 90 seconds to settle in to its stride. Also notice that fps can actually appear to be running higher earlier on in the test with lesser performing setups.

 

 

Delta is computed using the standard deviation formula:

 

 

In short, the EFSA setting provides more bandwidth between exe addons and P3D.

 

(Tests conducted on a 4.1GHz3960x+GTX680 AM=1360=01,01,01,01,00,00 1920x1200 monitor, DSR 1.5x full simulator conditions, overcast, Ai injection, shadows, high LOD and autogen).

I recommend turning on HT. Remember we're not splitting the power of each core into two. Rather, we're halving the task switching time within each core, but to do that the core looks like two cores to an app like P3D. So we need to help apps like P3D to understand that those two cores are actually only one core, presented as two logical processors.

 

We enable alternate LPs for the multi threaded apps like P3D, e.g. 85=01,01,01,01. These four unmasked LPs are allocated one per core. We do this so that when P3D starts up, it "sees" four cores, and breaks itself into four parts, one per core in this case.

 

Applications that move program affinity or alter priority should be avoided, as this can induce instability into the system. Batch files do the best job of starting apps externally with affinity since they use the Windows app starting functions directly, or apps that properly support the process affinity space (like IF10) can be used without problems.

 

So to the 8 core:

 

 

OK Mike. Let's look at what I would do with your 8 core.

 

Turn on HT

 

Set up an affinity mask in P3D jobscheduler section of AM=21760=01,01,01,01,00,00,00,00

 

I would run all addons on cores 0 and 1, AM=13=D=11,01 and AM=14=E=11,10, giving them each three LPs.

 

This leaves a pair of cores, 4 LPs on cores 2, and 3, which becomes a nice wide "hole" in the CPU utilisation for the jobscheduler to hit with other tasks.

 

This "hole" is set up in front of the P3D cores and is important.

 

I would set IF10 EFSA=True

 

I would start addon exe apps outside of exe.xml with a .bat or IF10.

 

If you're running PC's attaching from the physical network or don't have EFSA, move the unmasked four LPs toward the front by one core i.e AM=5440=00,01,01,01,01,00,00,00

Hi Steve,

 

Many thanks for the explanation regarding EFSA and your further recommendations. I'll get around to conducting further tests as soon as time permits. Chasing my tail right now with real world stuff and then there's that little matter of updating P3D to v3.2 All the preparation has been done so shouldn't present any problems.

 

Regards,

Mike

Mike, you definitely will want to update to V3.2 ... fixes a lot ... along with performance and AA improvements.

 

I was running a quick test last night out of Popham with Orbx EU/airport and ASN running, it was stunning with all the graphics options ON/Max.  Made a quick video for a friend who actually flies real world out of Popham and she was impressed at how close it looked to her RW experience.

 

I just finished "Catch-up" on all the recently release V3.x Orbx airport updates and tossed in a few new FSDG and Aerosoft airport ... the quality is getting so good now that I spend way too much time looking out the VC trying to "catch" all the little details ... ignoring what my aircraft is doing and where I'm going.  I'm in GA mode right now, I'll switch over to "commercial" mode once I get my fill and break open my TAM 2016 book and start reading on long flights.

 

Cheers, Rob.

Hi Rob,

 

Looks like you're having fun! Such interludes must be very welcome between all that testing.

 

They say patience is a great virtue, but sometimes it's damned hard to live up to it! I guess I'll be finding the time somehow to do the deed this weekend. Hopefully shouldn't take long.

 

Regards,

Mike

Yeah haven't flown much recently ... certainly not a good A to B flight I want to do ... I've done quick tests and the like, but would be nice to get back to a nice A to B flight with some flight planning.

 

Orbx have just updated ObjectFlow so that completes all my outstanding issues with V3.2.

 

Cheers, Rob.

Well, that's me up to v3.2!! I opted to install all 3 components (Client, Content and Scenery) and the whole process was remarkably straightforward without any problems. The Shaders folder was deleted prior to the update and System Restarts performed after uninstalling the 3 components and following the reinstallation of their v3.2 equivalents.

 

ObjectFlow was updated via FTX Central 2. FSUIPC produced a warning at the beginning stating that I was trying to install to a later version of P3D - just continued and it completed without any errors. Updated FlightSim Lab's Spotlights to v1.8. Reinstalled FTX Global BASE. Haven't touched 'REX Texture Direct with Soft Clouds' as waiting for their update. I ran REX on the off chance and was invited to install a Beta update but chose to decline at this stage. I'll wait for the official release.

 

I file-compared several files and folders using the nifty utility KDiff3 including terrain.cfg, autogen folder, effects folder, DLL.XML and Prepar3D.cfg and the only changes noted were the addition of a couple of files in the effects folder relating to the return of those waves and 4 ORBX Street Light entries had been removed. I ran the configurator to reinstate those entries, but it didn't make any difference so I suspect that these entries are probably irrelevant.

 

All in all a very pleasant experience. Prepar3D v3.2 (yeh!) ran first time with no errors. It remains as rock solid stable as its predecessor and I like the new Menu layout. Frame rates are good although not sure they are much better than that observed in v3.1. I turned up Shadow quality from Medium to High and Water from High to Ultra. Flew a couple of flights out of KSEA and KVPS and first impressions are very good (except for those default clouds - ugh!). The Prepar3D Profile was returned to NVIDIA defaults in NVIDIA Inspector and MSAA turned up to 8 Samples in P3D. Yes, again early impressions confirm that AA is much improved - so far I've tested with RealAir's Duke and Legacy as well as the default B58 and Raptor. It may be my imagination, although I don't believe so, but the lighting seems much better. I flew under building storm conditions and it was remarkable how the cloud shadowing affected the in-cockpit lighting - the rate of transition from light to shade and back again seemed very convincing and realistic.

 

However, I have the impression that while performance seems somewhat smoother the sim is still not altogether happy running on 4 of my CPU's 8 Cores. The Affinity Mask setting of 1360 remains in place and Hyperthreading is still enabled as the System remains in testing configuration. Formal testing is next! Watch this space, but please don't hold your breath as it may be a few days before I can devote some more time to producing some hard data. Don't worry, Steve, I've taken all your advice on board and fully intend to adhere to your very helpful guidance

 

Cheers!

Mike

Yes Mike, remember that AA functions use the GPU so turn off AA or at least down so that it does not cap fps. Turn off any limiting functions VSync or NI limiter. If you want to convince me you have made a good test, you need to show almost identical graphs. Graphs with big changes show a problem and are void.

 

Allocating five cores is never any good, 6 works a lot better almost as good as four on the 8 core. So also try AM=21840=01,01,01,01,01,01,00,00 - still leaves two cores ahead of those 6 cores for addons. But I think you'll find the four cores work best, unless you got something weird going on. As I remember, so far I've not seen a consistent pair of results in your graphs. Show me 6 cores working better than 4, and 8 cores working better than 6.

Hi Steve,

 

Okay. Thanks.

 

One question: If our results are to be compared, does it matter where the flight testing takes place? Would it be helpful if I ran a test flight using your settings - in sim settings, active scenery Addons, chosen aircraft, airport, altitude and heading? Would our system hardware differences have any significant affect on the resultant average deltas?

 

Actually, that was three questions, but they are related

 

Regards,

Mike

Choose a heavy sim situation or a light one, doesn't really make any difference Mike. Heavy situations can show up differences in setup more readily though. When testing for sheer performance, that is ultimate throughput, you'll see that with the unlimited setting and no AA. You want to capture progression of that throughput in your graphs. You want to see 6 cores working better than four - or whatever. Then when your system has the ultimate throughput capability, set up NI limit and Vsync + an acceptable amount of AA.

 

To be honest, I'd like to see you able to demonstrate the advantage of HT enabled, that would be good.

the sim is still not altogether happy running on 4 of my CPU's 8 Cores.

Thinking out loud Mike, when you say that, to me it suggests logically there's something badly wrong in there. I hope that gets sorted out because it would be a shame to go back to HT off no AM.

 

 

 

Reason the four core guys generally get a better experience is because they've less LPs in the first place to mount up over each other.

 

8 cores are useful with respect to P3D (or FSX), those surplus four cores soak up all the other activity to produce a better result than six cores, and certainly four.

 

Testing P3D v3.2 recently on a 4GHz class CPU, showed five jobs still a poor result, but six almost as good as four, but eight none better than six. So why allow more than four jobs, since they only spawn 50 or more threads and they invoke activity across the CPU through the system resources. You don't want excess P3D jobs filling those holes. I found six cores v3.2 better than 6 cores v3.1. So maybe with a mega system in the near future 6 cores may be better than 4.

Hi Steve,

 

I realise that I'm not comparing apples with apples, strictly speaking, but this 'situation' only came about because of my long time interest in Flight Simulation and my conviction that Prepar3D rises head and shoulders above all the rest, hence the reason why I have it on my drive.

 

I have mentioned elsewhere that several other popular resource hungry titles also reside on the same SSD 1TB drive: GTA5, Project Cars, Elite: Dangerous/Horizons, X-Rebirth, Skyrim (heavily modded). Heck, I even have Microsoft Flight which I enjoy firing up from time to time. The point I'm trying to make is that each and every one of those titles performs beautifully without a trace of stuttering of any magnitutude.

 

If Prepar3D had not been installed, the likelihood is that my i7-5960X would be running with Hyperthreading enabled and I would have been none the wiser. I have no idea whether any of the above use Affinity Mask tricks to optimise their performance, but I suspect they do not. I imagine this is simply because the balance favours the GPU, unlike Prepar3D - although that side of things is definitely improving. This theory is supported by the fact that I see no difference in performance with these other titles while Hyperthreading is disabled.

 

They say ignorance is bliss and, were it not for Prepar3D, none of this research and experimentation would have occurred. Running Prepar3D with CPU Hyperthreading enabled was okay but, at times, the microstuttering became unacceptable. Happily this all went away when I read that Rob had discovered that running on all 8 Cores without Hyperthreading was delivering much better performance. I do understand your sincerity when you keep stating that 4 Cores with Hyperthreading enabled should, logically, be the desired option and if it isn't then 'there's something badly wrong'. But what? If you are correct, and I'm not saying that you are not, then that would imply that Rob's setup, which is similar to mine, would also be defective in some way were he to use a similar configuration. This does seem very unlikely, wouldn't you agree?

 

Don't get me wrong, this whole process has proved to be quite fascinating and I remain willing to do what it takes to produce any necessary evidence proving which is the best way to go on my system. For the rest of you who may be reading this, RELAX! All the agonising in this thread over how to achieve best performance with Prepar3D only really applies to those who may be using an 8-Cored CPU, but then you knew that, didn't you? :-)

 

Regards,

Mike

Don't worry if you cant figure it out, set what suits you Mike. But if you're doing tests and making graphs and comparisons, they need to make sense like the ones i'm posting. The poor setups gravitate to HT off no AM and there's a scientific, mathematic reason for that. You should be able to prove HT on is superior, but that takes a little know-how.

The poor setups gravitate to HT off no AM

 

So, Steve, can you provide some examples of what would qualify, as you see it, as being a 'poor setup'? Sorry to keep prodding but, and perhaps inevitably, once again you have invited my response :-)

 

The balance of probabilities suggests that my setup will end up being categorised as 'poor' and I would like to know why. Previously you have turned this around by making it my responsibility to find out why it is falling short of optimal performance but, without any pointers, it's difficult to know where to start since my instincts tell me that the system/setup is performing well.

 

For the purposes of testing, I can...

1. Turn off Antialiasing, VSync and Triple Buffering, run Unlimited and leave NVIDIA Inspector out if the equation.

2. Ensure other programs such as those that restrict traffic and inject weather start with their own Affinity Mask by using batch files. (IF10 Pro is a special case and, like Prepar3D, the Affinity Mask can be set in their respective configuration files. Also, IF10 Pro can be used to inject weather.)

3. Disable Process Lasso, including the core engine, thus preventing any ProBalance activity and consequently no priority adjustments.

4. ESET Nod 32 Antivirus has a gaming mode which can be turned ON for the duration of the tests. However, this exposes the system to vulnerabilities so probably undesirable.

5. I could run with or without SLI enabled although, understandably, I prefer to run enabled.

 

What more can I do? Windows 7 64bit Pro is rock solid stable and continues to be very responsive since the day of installation a few months ago. Windows Update is run manually every month or so. There are no active processes that might impede performance while Prepar3D is running and those that do raise their heads will be managed effectively by Windows. Hardware? Again no problems and MoBo BIOS is current with all drivers in place and functioning correctly. Current GPU driver: 361.91 WHQL and no conflicting audio drivers are installed.

 

You should be able to prove HT on is superior, but that takes a little know-how.

Sounds like I may be on a hiding to nothing since I am under no illusions regarding my capabilities..LOL!

 

Cheers,

Mike