500MB VAS reduction and increase FPS by 22% in V3 ... need some testers

[Chris Bell 165]

It's the same for 2.5

Thank you Steve,

if that’s the case; we now know this isn’t a P3D v3 specific bug,

or a side effect from overhauling VAS management in v3!

[331BK 96]

Hmmm i will test it tommorow.

 

But it can match with my problem with vas.

 

Even with normal p3d settings p3d does not release much vas.

 

I tested with pmdg777 from eddp default scenery to egpn with asn ,utx europe and gep3d.

 

So 1.5 hours flight in the beginning 1.2 gb vas free in the end 120 mb.

 

I use a 5820k @ 4.5 ghz with ht on and no am. I will try the same tommorow with ht off.

 

I'm verry intersted to see if there is a difference

[Guest JustanotherPilot]

In Prepar3D v3 I'm getting a smoother performance with HT on and AM 170, than with HT off (AM tried on and off) using the i7-4790K(no OC), VAS seems to be the slightly better with HT off, but nothing to warrant keeping it turned off.

 

My findings are that it seems better with HT and AM active than no HT or AM.

[Guest]

Anyone else tried this with a 5960X ?

 

Going back to my 3960X days and P3D V2.x ... I had done this video back in Jan/2015 with HT ON and setting Affinity on the fly ... you can see the effects of turning ON/OFF cores so I know it was working.  My conclusion from 6/12 3960X was to NOT use AF (defaults to all cores) and leave HT ON ... which is contrary to Steve's information.

 

 

With the 5960X and V3 this is not the desired approach ... I'll need to do more testing and I never really did extensive testing of VAS with HT/Affinity in 3960X.  But again the unexpected surprise is the consistent 450-500MB of VAS I'm saving on every flight so far with HT OFF.

 

I'm not ruling out any possibilities at the moment (OS, CPU, C++ libraries, LM code), but it almost looks like duplicate threads are being created when "physical" cores exceeds 4.  But without access the source code and development environment, it would be very difficult for me to positively validate or invalidate my theory.

 

Cheers, Rob.

[Chris Bell 165]

some boards bios allows you to limit cpu cores;

you could try to limit to 4 cores and see what happens,

 

if the theory is true; you should technically get even more performance when limited to 4 cores with HT on!

[DylanM 1,069]

^Was just going to suggest that...! My Gigabyte Gaming5 has the option to turn off specific cores - will have to give it a try.

[S_D 7]

Is there any (free!) software I can use to record flights so they can be replayed over and over with different BIOS settings? I'm not talking about recording the video, but the actual 'flightpath' as it were....

[be77solo 125]

Only stutters with the Q400 are when viewing from "Spot" view ... don't use that view and used "Locked Spot", otherwise performance is excellent in the Q400.

 

Great, thanks for the tip, "Locked Spot" does solve the problem!  I had avoided the Q400 since the upgrade, but happy to be flying her again!

 

Seems from reading this thread I need to investigate AM further...

[Chris Bell 165]

fs recorder can record your flight and play it over without video recording it

[Noel 4,519]

Sounds entirely predicted w/o resorting to a duplicate threads theory.  Isn't the primary issue re additional cores ever since FSX SP2 [or 1, I don't recall] that those cores are utilized for terrain texture loading?  If more terrain textures can be loaded in the same interval of runtime then it arguably follows that VAS use goes up per interval of runtime.  Because of this one would guess the same effect happens w/ V2.x as was discovered by the OP's theory.  The 'increased [single-core] performance' has also been well known when HT is off versus on.  In part from lower total vcore need and therefore lower heat associated w/ HT off versus on.  I know while frame rate performance may increase w/ HT off, HT on has always yielded best 'total performance', i.e. image quality, texture update rate, freedom from blurred textures, etc on my SB-E machine.

[swiesma 625]

Seems to me like a problem with more then 4 physical cores.

 

Will test tonight after work switching off HT on my i7 6700k @4,5 GHz.

[Mickeyj 35]

G´day Rob,

 

as promissed her are my test results. I also tried AM=1364 based on this calculator http://www.gatwick-fsg.org.uk/affinitymask.aspx?SubMenuItem=hardware

Results were similar to HT OFF

 

Test machine:

 

Intel i7 5820k oc´ed 4.5GHz +++ ASUS Rampage Extreme V MoBo +++ GTX 980 SLI +++ 16GB Ram +++ Win 64 bit PRO dedicated SSD +++ P3Dv3 dedicated SSD

 

Scenario:

 

Carenado Hawker 850XP Lite running engines

Recorded flight from KPAO (Cielosim) - KSQL (ORBX) - KSFO HD (Flightbeam) base scenery ORBX FTX NCA ASN Weather REX Soft Clouds

Please note FTX NCA + ORBX KSQL are not officially supported!!!

 

 

FSUIPC Log:

 

HT OFF:

 

********* FSUIPC4, Version 4.946b by Pete Dowson *********

   174799 E:\Program Files (x86)\Lockheed Martin\Prepar3D v3\SimObjects\Airplanes\Carenado H25B_H850XP_LITE\H850XP.air

   174799 C:\Users\MJ\AppData\Roaming\Lockheed Martin\Prepar3D v3\Temp\RECTMPFLT.fxml

   175751 Monitor IPC:024C (S8) = 44

   175782 Aircraft="CARENADO H850XP N5847Z LITE"

   185797 Memory in use: 3398Mb, Avail=698Mb

 

   246138 Memory in use: 2954Mb, Avail=1142Mb

 

   306433 Memory in use: 3047Mb, Avail=1049Mb

   

   366805 Memory in use: 3131Mb, Avail=965Mb

   

   427037 Memory in use: 3196Mb, Avail=900Mb

   

   487347 Memory in use: 3184Mb, Avail=912Mb

   

   547579 Memory in use: 3240Mb, Avail=856Mb

   

   607827 Memory in use: 3236Mb, Avail=860Mb

   

   668043 Memory in use: 3236Mb, Avail=859Mb

   

   728244 Memory in use: 3259Mb, Avail=837Mb

   

   788429 Memory in use: 3165Mb, Avail=931Mb

   

   848661 Memory in use: 3169Mb, Avail=927Mb

   

   908956 Memory in use: 3188Mb, Avail=908Mb

   

   969141 Memory in use: 3221Mb, Avail=875Mb

 

 

HT ON:

 

********* FSUIPC4, Version 4.946b by Pete Dowson *********

   156282 E:\Program Files (x86)\Lockheed Martin\Prepar3D v3\SimObjects\Airplanes\Carenado H25B_H850XP_LITE\H850XP.air

   156282 C:\Users\MJ\AppData\Roaming\Lockheed Martin\Prepar3D v3\Temp\RECTMPFLT.fxml

   157280 Aircraft="CARENADO H850XP N5847Z LITE"

 

   258135 Memory in use: 3081Mb, Avail=1015Mb

 

   318274 Memory in use: 3068Mb, Avail=1028Mb

 

   378568 Memory in use: 3221Mb, Avail=875Mb

 

   438737 Memory in use: 3247Mb, Avail=849Mb

 

   499016 Memory in use: 3332Mb, Avail=763Mb

 

   559186 Memory in use: 3385Mb, Avail=711Mb

 

   619449 Memory in use: 3352Mb, Avail=744Mb

 

   679634 Memory in use: 3374Mb, Avail=722Mb

 

   739975 Memory in use: 3279Mb, Avail=817Mb

 

   800254 Memory in use: 3288Mb, Avail=808Mb

 

   860533 Memory in use: 3307Mb, Avail=789Mb

 

   920734 Memory in use: 3341Mb, Avail=755Mb

 

 

Settings:

 

 

 

 

 

 

Video HT OFF KSFO flyby

 

 

Video HT ON KSFO flyby

 

[S_D 7]

Seems to me like a problem with more then 4 physical cores.

 

Will test tonight after work switching off HT on my i7 6700k @4,5 GHz.

 

Maybe the answer is to ask someone to test on a hyperthreaded i3 (2C, 4T) to see if performance improves with HT Off.

 

@Rob, do you know of a way I can 'record' the flights so I can reproduce them identically after changing my BIOS setting?

[SteveW 1,792]

Maybe the answer is to ask someone to test on a hyperthreaded i3 (2C, 4T) to see if performance improves with HT Off.

A dual core is a special case and should be used with HT enabled for FSX and P3D, and with one of the first LPs masked. A dual core with HT off will put the third thread on the same core as the main thread, with the second thread (least busy) on a core to itself.

 

If we get an improvement with HT disabled, that would mean our HT enabled Affinity was set incorrectly.

 

To reiterate; With HT enabled we need to take care of that first thread by masking one of the LPs, since it won't be up to speed when the Jobscheduler assigns the next LP, and will appear on the next available LP, namely the one on the same core. The second thread isn't doing a lot but can take upward of 8% throughput from that core. We can limit the number of LPs to around 6 to save too many threads gathering data, generally these won't be maxed like the main thread.

[Guest]

 

@Rob, do you know of a way I can 'record' the flights so I can reproduce them identically after changing my BIOS setting?

 

Yes, P3D V3 has a built in option to record your flight and save it so you can play the exact same flight back.  This will work for VAS testing, but it will NOT work reliably for FPS testing ... the playback system seems to have a case of jitters so FPS numbers will not be reliable.  But if you are going to use the built in recording facility then you should set AI to 0 and remove any weather (clear skies) to avoid possible variance/skew.

 

A good way to record FPS is use the same concept that FSXMark11 does ... load an aircraft with AP setup, save flight.  Then load flight and just let it run the AP course then log with Fraps time frame facilities and/or use FSUIPC logging.  But again you really should use clear skies and disable AI ... you can test with AI and weather engine enabled using same archived date/time but there will always be some variance even with exact same time/date used.

 

The built recorder system will exclude ability to test ASN and AI traffic VAS influence but can be used for testing out scenery/location and airports VAS usage.

 

Cheers, Rob.