No really, I broke it, but it does work great now!

 

Caveat: What I did is probably wholly useless for the high-flying commercial pilots - this is strictly a GA environment with Orbx terrain and textures.

 

First, some background. I purchased P3D V2.4 for my Christmas gift and have been fighting it for the last three weeks to improve terrain textures. I fly strictly GA, maybe a bizjet or two, and find that ground textures, clarity, and crisp autogen are paramount to my flying experience. After all, if I were in a real GA airplane I would not be fiddling with the FMS or other distractions. Nope, I would be scanning the sky, but equally importantly, scanning the ground. I want to look down and see (in an ideal world) the guy mowing his lawn, the different neighborhoods, people playing on the beach, etc. While this level of detail is impossible outside of an Orbx airport, I really find it distracting to have blurred textures when I am flying. If they resolve themselves at LOD radius = 1 right under the plane, so be it - as long as they resolve! Jitters are another bugaboo, but not as bothersome as blurries to my eyes. 

 

Flying in a WACO, C337, or the A2A C182T does not usually pose a problem: they are slow enough that the terrain resolves itself. In fact, flying the C182T was the thing that really opened my eyes to how much detail I was missing in other, faster aircraft! The terrain clarity is stunning while flying that bird, a combination of a great model and relatively low speed. However, as the speeds ramped up, the terrain begins to blur. This is particularly true for the Legacy, which is a little hot-rod in a class of its own for combining sight-seeing visibility and effortless speed at tree-top levels.

 

After trying a TON of different things, I accidentally introduced a combination of features that essentially "broke" the P3D V2.4 CPU consumption model, such that NO SINGLE CPU IS SATURATED and therefore presenting a bottleneck.

 

Through a combination of enabling hyper-threading (which I always found to be a detriment in FSX and P3D V1.4), assigning affinity to system resources but having a mismatch on CPU affinity in Prepar3D.CFG, and really cranking up FFTF and SWT, I have a system that allows me to fly the Legacy over Portland, OR (Orbx PNW) at 200 knots and 2,500 feet with no blurries, very few jitters, smooth turns, and while FPS dips into the low 20's, everything is smooth. I have never been much of frame hound, as long as it is smooooooth.

 

Here is how I did it.

 

I recently picked up Process Lasso, a terrific little program that allows you to easily slice and dice CPU affinity and priorities. While my first efforts were rewarding (move everything that was not part of the flight sim ecosystem to CPU 0), I was still frustrated, as virtual core 0 for P3D was hammered and clearly representing a bottleneck. Remembering OZwhitey and his experiments with the 6-core 5280, I turned on hyper-threading and, using Process Lasso, assigned affinity for P3D to threads 2-7, leaving threads 1-2 to system activities and Opus ESI. Oh, and I also clicked the box that said "This is a game" based upon the described benefits. Yeah, I know - it hurts... ^_^  This worked great, and there was a massive improvement in smoothness and visual clarity. Wait a minute - I had failed to change the affinity mask in my Prepar3d.CFG file and it was left at Affinity=14, i.e., cores 1-3 in a non-hyper-threaded 4 core processor. I checked Task Manager, which I always leave running during testing to monitor CPU and VAS consumption, and to my utter surprise NONE of the six active threads hosting P3D was hammered! Whoa! They weren't even, with maybe the highest at 60-percent and the lowest at 30-percent, but again, none of them was completely saturated.

 

Conducting a short test by commenting out my affinity mask setting in my CFG file showed that P3D behavior reverted back to normal, i.e., virtual CPU 0 was hammered again. Crossing my fingers, I reinstated the affinity mask and we were back to all six threads active, but none of them hammered. Magic or voodoo, not sure which, but it works! I think part of it is the Process Lasso "game" setting, but have yet to test what happens when I turn it off.

 

This opened the skies (pardon the pun)!  I now had sufficient CPU cycles to improve terrain texture behavior, and through a bit of reading and trial and error bucket tests (use a small bucket - see what happens - use a big bucket - see what happens, retry until the optimal bucket size is reached, i.e., a binary sort) I arrived at FIBER_FRAME_TIME_FRACTION=0.50 and SWAP_WAIT_TIMEOUT=50. Now, I know there has been some chatter on AVSIM about setting FFTF=0.01 and people reporting good frame rates, but that, at least to me, was voodoo. I have riding fast motorcycles long enough to know that "more" is better than "less" (smoothness not withstanding), and you don't get to go fast unless you add more. After all, the purpose of FFTF is to provide enough CPU cycles to process the terrain textures. The other variable is SWT. My understanding is that this is the amount of time the sim will wait, after which it assumes that terrain processing has been completed. Together, they seem to represent the primary "knobs" for tuning terrain processing.

 

After putting this combination in place (and yes, there was a LOT of serendipity involved in the whole process), I have a sim that allows me to fly the Legacy over heavily populated Vancouver, WA or Portland, OR at 2,500 feet at 200 knots and have fully resolved terrain under me. Sometime it resolves late, sometimes early, but it is never falls behind and turns into a blurry mess. Even flying into Orbx KTVL, which was just brutal on my P3D V1.4 installation, is smooth and unblurred. Just for grins, I decreased altitude to 1,500 feet above Portland @ 200 knots, and while the terrain was fully resolved I was getting a wee bit too many jitters, suggesting that some other P3D process was being starved for resources.

 

There is still some tuning to go, that's for sure, but I feel much better having a "broken" foundation under my wings that responds, in a deterministic manner, to input! Oh, and trying different affinity masks in the CFG file? Ain't gonna happen...    Fourteen is just fine.

 

Finally, just to show you how weird things can get, I had micro-stutters after my "fix" that were just distracting. Turns out the combination of my GTX 780 and 2560 x 1440 ASUS monitor is incapable of holding down 60 hertz, but was only producing 59. Using NI I changed to match and viola, no more micro stutters.

 

Isn't tweaking fun?!?!?!

 

Here are my CFG settings if anybody wants to comment/copy. I put all the "tweaks" up top so they are easily accessible, and this is not the entire CFG file. Note that there are still some compromises - my LOD radius is down from 6.5 to 5.5, and scenery complexity is "normal". However, Orbx building and tree density seems to be uncoupled from general scenery complexity, so I can keep those high. Love my vegetation!   

//
// CFG file constructed on 2015-01-19
// Variables that determine performance are set up front
//
[JOBSCHEDULER]
AffinityMask=14
[BUFFERPOOLS]
Bufferpools=0
USEPOOLS=0
Poolsize=0
//
//
[Display]
TEXTURE_FILTERING=16
MSAA=8
SSAA=0
VSYNC=1
FXAA=0
TRIPLE_BUFFER=1
SKINNED_ANIMATIONS=1
TEXTURE_BANDWIDTH_MULT=40
UPPER_FRAMERATE_LIMIT=30
TEXTUREMAXLOAD=27
WideViewAspect=True
FullScreen=True
FullScreenExclusive=True
FullScreenAutoFill=True
ChangeTime=4.000000
TransitionTime=4.000000
ActiveWindowTitleTextColor=255,255,255
ActiveWindowTitleBackGroundColor=0,28,140,64
NonActiveWindowTitleTextColor=255,255,255
NonActiveWindowTitleBackGroundColor=24,33,87,64
InfoUpperRightTextColor=255,0,0
InfoUpperRightBackGroundColor=0,0,0,0
InfoLowerLeftTextColor=255,255,255
InfoLowerLeftBackGroundColor=255,0,0,128
InfoLowerRightTextColor=255,255,255
InfoLowerRightBackGroundColor=255,0,0,128
InfoBrakesEnable=True
InfoParkingBrakesEnable=True
InfoPauseEnable=True
InfoSlewEnable=True
InfoStallEnable=True
InfoOverspeedEnable=True
InfoWeaponSystemEnable=True
InfoCountermeasureEnable=True
//
//
[SCENERY]
POP_FREE_AUTOGEN=1
AUTOGEN_BATCH_LOD=2
AUTOGEN_TREE_MAX_DRAW_DISTANCE=11000.000000
// // from 9500, max 12000
AUTOGEN_TREE_MIN_DISTANCE_TO_LOD=5500.000000
// // from 2500, max 6500
IMAGE_COMPLEXITY=2
//
//
[TERRAIN]
SWAP_WAIT_TIMEOUT=50   // from 30
LOD_RADIUS=5.500000
TESSELLATION_FACTOR=90
MESH_RESOLUTION=23
TEXTURE_RESOLUTION=28
AUTOGEN_VEGETATION_DENSITY=4
AUTOGEN_BUILDING_DENSITY=4
DETAIL_TEXTURE=1
WATER_REFLECT_CLOUDS=0
WATER_REFLECT_USERAIRCRAFT=0
WATER_REFLECT_SIMOBJECTS=0
WATER_REFLECT_TERRAIN=1
WATER_REFLECT_AUTOGEN_VEGETATION=0
WATER_REFLECT_AUTOGEN_BUILDINGS=1
WATER_REFRACT_CLOUDS=0
WATER_REFRACT_USERAIRCRAFT=0
WATER_REFRACT_SIMOBJECTS=0
WATER_REFRACT_TERRAIN=0
WATER_REFRACT_AUTOGEN_VEGETATION=0
WATER_REFRACT_AUTOGEN_BUILDINGS=0
ENABLE_BATHYMETRY=0
WATER_CLARITY=30
WATER_DETAIL=3
//
//
[DISPLAY.Device.NVIDIA GeForce GTX 780.0.0]
Mode=2560x1440x32
Anisotropic=1
//
//
[Main]
AlwaysFullLoad=1
FIBER_FRAME_TIME_FRACTION=0.50   // from 0.33
User Objects=Airplane, Helicopter, Submersible, ExternalSim
SimObjectPaths.0=SimObjects\Airplanes
SimObjectPaths.1=SimObjects\Rotorcraft
SimObjectPaths.2=SimObjects\GroundVehicles
SimObjectPaths.3=SimObjects\Boats
SimObjectPaths.4=SimObjects\Animals
SimObjectPaths.5=SimObjects\Misc
SimObjectPaths.6=SimObjects\Submersible
SimObjectPaths.7=SimObjects\Weapons
SimObjectPaths.8=SimObjects\Countermeasures
LabelsSupported=Airplane, Helicopter
LocationFullScreen=-201,0,2359,1440,\\.\DISPLAY1
Maximized=1
HideMenuNormal=0
HideMenuFullscreen=1
//
//
[GRAPHICS]
SHADER_CACHE_VERSION=1
MIPMAP_VC_PANELS=1
TESSELLATION=1
TEXTURE_MAX_LOAD=1024
NUM_LIGHTS=4
COCKPIT_HIGH_LOD=1
AIRCRAFT_REFLECTIONS=1
LANDING_LIGHTS=1
DAWN_DUSK_SMOOTHING=1
LENSFLARE=0
HDR=0
BLOOM_LEVEL=2
SHADOW_QUALITY=2
TERRAIN_SHADOW_CAST_DISTANCE=0
CLOUD_SHADOW_CAST_DISTANCE=3
OBJECT_SHADOW_CAST_DISTANCE=1
CONTENT_NO_SHADOW=0
INTERIOR_SHADOWS_CAST=0
EXTERIOR_SHADOWS_CAST=1
SIMOBJECT_SHADOWS_CAST=0
TERRAIN_SHADOWS_CAST=0
VEGETATION_SHADOWS_CAST=0
BUILDING_SHADOWS_CAST=0
CLOUD_SHADOWS_CAST=1
INTERIOR_SHADOWS_RECEIVE=1
EXTERIOR_SHADOWS_RECEIVE=1
SIMOBJECT_SHADOWS_RECEIVE=0
TERRAIN_SHADOWS_RECEIVE=1
VEGETATION_SHADOWS_RECEIVE=0
BUILDING_SHADOWS_RECEIVE=0
SHADOW_NUM_CASCADES_LOW=4
SHADOW_NUM_CASCADES_MID=5
SHADOW_NUM_CASCADES_HIGH=7
SHADOW_NUM_CASCADES_ULTRA=12
GROUND_SHADOW_TEXTURE_SIZE=2048
CLOUD_SHADOW_TEXTURE_SIZE=512
SHADOW_LOG_PARTITION_LOW=0.850000
SHADOW_LOG_PARTITION_MID=0.850000
SHADOW_LOG_PARTITION_HIGH=0.850000
SHADOW_LOG_PARTITION_ULTRA=0.850000
EFFECTS_QUALITY=2
EFFECTS_DISTANCE=2
TEXTURE_QUALITY=3
IMAGE_QUALITY=1
See_Self=1
Text_Scroll=1

Edited January 20, 201511 yr by n4gix
Moved .cfg entries to Code box for clarity.

I presume you're talking about the review a Lancer legacy… Yes?

 

Great tweaking....and migraine...

 

Chas

Heya Chas, 

 

Yes, to be specific, the RealAir Lancair Legacy. And yes, I do have a head-ache, compounded by a ferocious cold...   :(

 

Addendum!

 

I just flew from Mammoth (KMMH) to Redding (KRDD) over the top of Blue Mountain (KBLU) @ 26,000 feet in the Phenom, average speed (TCAS) about 350 knots. It highlighted something that I neglected to think about in my quest for low and slow: the LOD radius is like a globe (force field?) around the aircraft, so the higher you are the nearer the intersection of that globe is with the ground, i.e., terrain that should be relatively sharp because it is not that far away as the crow flies is not because of the added height dimension. It is only when I started my descent that distant objects/terrain became sharp. And that was one really fine approach into KRDD - sharp as a tack!

 

Never end, does it?

Thanks for posting John, this may help others who are having blurry issues.

 

Cheers, Rob.

Thanks, Rob.

 

I am curious if anybody can replicate what happens with the mismatched processor affinity assignments. Process Lasso is free for 7 days, turns to nag-ware for the remainder of the month, so the test would not be expensive.

What kind of processor do you have ?

Do you have any extra programs running and if so ob which cores ?

 

I have a 5820k and am curious If I can gain something from this.

I have paid for Process Lasso and have tried various combinations but I ultimately ended back with NOT setting any affinity value ... so in my case I was not able to replicate, but I'm using 6/12 core 3960X.

 

Personally, I would never turn off HT in any processor released within the last 5 years or so.  Intel did have some early issues with HT but those days are long gone.

 

Of course the OS will be a big part in the threading picture ... Microsoft OS's have improved their threading efficiency over time i.e. Win8 is better for threading than Win7 is better than WinXP ... so your OS of choice will also impact results.  Although I don't use Win8.x much because of it's UI and breaking changes that weren't necessary other than for marketing reasons, it does provide the best threading environment along the MS line of OS's.  

 

Cheers, Rob.

And...don't forget dx12, Rob

 

Chas

What kind of processor do you have ?

Do you have any extra programs running and if so ob which cores ?

 

I have a 5820k and am curious If I can gain something from this.

 

Hi Gerard, 

 

I have a 3770K.

 

I have normal anti-virus functions active (afterall, I am still connected to the Internet) and run Opus FSI weather, all of which I have confined to CPU 0-1. There are some exceptions that Process Lasso warns about, Explorer being one of them, so I left them alone. Process Lasso does a pretty good job at identifying which processes are using how much CPU at a given time and cumulatively. For example, I use my PC to charge my phone, and that seems to use a lot of cycles. I confined that application to CPU 0.

 

I have paid for Process Lasso and have tried various combinations but I ultimately ended back with NOT setting any affinity value ... so in my case I was not able to replicate, but I'm using 6/12 core 3960X.

 

Personally, I would never turn off HT in any processor released within the last 5 years or so.  Intel did have some early issues with HT but those days are long gone.

 

Of course the OS will be a big part in the threading picture ... Microsoft OS's have improved their threading efficiency over time i.e. Win8 is better for threading than Win7 is better than WinXP ... so your OS of choice will also impact results.  Although I don't use Win8.x much because of it's UI and breaking changes that weren't necessary other than for marketing reasons, it does provide the best threading environment along the MS line of OS's.  

 

Cheers, Rob.

 

Heya Rob, 

 

Empirically, I found that hyper-threading on FSX and PS3 V1.4 did not provide any benefit, but things change - somewhere along the lines a service pack was introduced and things got better.

 

Given your setup, do you still see core 0 being hammered in your flight simulator? That is the SINGLE most significant change that I stumbled upon - the fact that no CPU is being hammered into being a bottleneck.

Given your setup, do you still see core 0 being hammered in your flight simulator? That is the SINGLE most significant change that I stumbled upon - the fact that no CPU is being hammered into being a bottleneck.

Hi,

 

this very effect can be remedied quite easily. Start the simulator, go into Task Manager, switch off core 0, close TM, open it again and re-activate core 0. No Process Lasso required (which I own, too, but don't use at present).

 

This is a well-known trick, and it has been discussed on the LM forum. Someone from the LM team (Zach Helmuyn?) advised against it, though, as it would destroy P3D's natural threading mechanism. According to him it's inevitable to have core 0 hammered, at present.

 

I tried the core-0-trick, and it works (in the sense of  equalizing core load), but doens't help anything visually noticeable in my case (system below), thus I usually leave it alone.

 

Kind regards, Michael

Do you run you processor overclocked ?

 

Tomorrow I am going to try the demo and see what it does for my system.

Hi,

 

this very effect can be remedied quite easily. Start the simulator, go into Task Manager, switch off core 0, close TM, open it again and re-activate core 0. No Process Lasso required (which I own, too, but don't use at present).

 

This is a well-known trick, and it has been discussed on the LM forum. Someone from the LM team (Zach Helmuyn?) advised against it, though, as it would destroy P3D's natural threading mechanism. According to him it's inevitable to have core 0 hammered, at present.

 

I tried the core-0-trick, and it works (in the sense of  equalizing core load), but doens't help anything visually noticeable in my case (system below), thus I usually leave it alone.

 

Kind regards, Michael

 

Heya Michael, 

 

Interesting news - I had not seen that before. I will retain my current settings and continue to explore to see if the "natural threading mechanism" is upset. Shouldn't, since a thread is a thread, and it is only becomes an issue when you wait for other threads to complete (i.e., a spin lock is about as bad as you can get). Every flight so far has been smooth, blurry-free, but then again I only found this fix two days ago.

 

 

 

Do you run you processor overclocked ?

 

Hi Gerard.

 

Yes. My MSI system board has a button on it called "Overclock Genie", and when engaged it boosts performance from 3.5GHz to 4.2GHz. I was manually overclocking at 4.6GHz at one time, but the system became unstable so I just use the Genie.

John,

 

I found the topic on the LM forum (via google - search in that forum is a pain)

 

http://www.prepar3d.com/forum-5/topic/22fps-but-gpu-on-45-and-cpu-40/

 

Contribution #50009 by Zach Heylmun (sorry for misspelling his name).

 

Certainly a thread is a thread, but as far as I understand the point is co-ordinating them. 

 

Best regards, Michael

 

 

(via google - search in that forum is a pain)

 

Hah! True of most forums, it seems!

 

Thanks for the link, Michael - will check it out. Still playing with FFTF and SWT values.

Given your setup, do you still see core 0 being hammered in your flight simulator?

 

This is what my setup is doing ... ran some quick tests using various core configurations.  For my setup the default ALL cores seems to work best but still very minor difference 1-2 fps when using ALL cores. 

 

 

Cheers, Rob.

 

EDIT: you'll need to watch in 1080 HD full screen to see the text/values.