My experiences with the EVGA 780Ti 3GB (vs 670FTW) and 5760x1200

[theskyisthelimit 48]

Yeah I didn't give details since it was just a "what if" flight. Funny the 7950 ran with those settings and worked great but the 780 oomed in minutes.

 

I actually changed a bunch of stuff since then so no idea what the settings where. Pretty much line for line of Kostas settings.

 

Airfield was Orbx - PAKT, PFJ, Full rain (realtime weather with Opus and REX) night time. A2A 172 - no GPS. LINDA, SPAD, RemoteFlight, VAC, GoFlight pushing 5 panels 2 throttles, yoke, pedals, Cessna Trim Wheel.

 

I am going out tonight so will check again on Sunday or maybe tomorrow night.

 

I am doing my usual wait and see on P3D2. If I can not get my Saitek panels working in it I will not be onboard.

 

On p3d.. they give you a 60 day refund period.. which is kinda nice

[theskyisthelimit 48]

Its worth noting that with 4SS my max memory is around 2040 and GPU % around 59%.. if i switch to 4SGSS this becomes 2500MB and 99% (bufferpools = 0).. this in the FTX PNW region (with full on clouds and rain), this is on the 780ti 3GB (max volts shows 1.2v), while core max 1137 and memory 1749, temp max 74C, fan 79%. (external lock at 30)  

 

Only if you add clouds with rain does it hit 99% on 4xSGSS.. otherwise around 66%. (4xSS with weather and rain and a heavy region gpu load is 84%)

 

I would think that the fact the gpu hits 99% means either i should tweak fsx.cfg differently than maybe UsePools=0  or maybe sli of 770 or 780 would actually be of help.. though a recent test with scenery complexity down a notch still resulted in 22 fps to 26 fps in severe KSEA weather at 4SGSS.

 

I'm also hoping moving to the haswell 4770k at 4.5 or higher will free up any CPU bottleneck that could also be occurring, though i'm guessing even this 780ti cant keep up with the data being sent.. hence 99%.

 

**Its interesting to note that even lowering to 3840x800 and 4SGSS, utilization still hits 99% (3840x800-1050 is where most that have 4SGSS seem to fly on with good success).. frame rates go from 11.05 to 22 fps or so, for the Word Not Allowed Cloud test AA (too grainy to use on my screens though, non native).  So utilization maybe not a big indicator of much here.

 

 

Updating here.. 

 

I've gone from haswell (delidded 2 cpus in fact).. all the way backward a little to 4930k (at 4.4ghz with 2400mhz ram) and quite happy with the 6 core/12 HT choice.. hard drive loads for any app are greatly improved, including fsx.. only a slight dip compared to haswell in fps in fsx, maybe 2-3 at best.. so a good compromise  (this particular test case below, with the 4770k at 4.4 i was around 18.46-18.80, so it only went down slightly for more intense tests)  Also.. i had the 2600k at 4.6 before all this, with the other fsmark07 test i think it was ringing in around 27 fps all things held equal and now around 32 max for that particular test.

 

However.. even with the 780ti and the good hardware, 4SGSS (sparse grid) is still a killer at 5760 x 1080 (i switched to three new acer 3d VG248QE monitors).. I run fsx at 36 fps int. locked with NI at 1/4 refresh

 

Take my test case.. if anyone wants to test feel free to compare results.. 

 

I have this flight at this location It could be downloaded and tried to compare..

 

Basically its rudimentary, not as "automatic" as fsmark07 but more telling of performance..

 

just launch it, use fraps, set to a 60 second timer in fraps.. fraps set to p for benchmark key.. make sure joystick/yoke not turned but throttle at max on load.. plane should fly straight ahead, a little to right at first, then it slightly turns left towards end.. look at results.. thats my 18~ fps.  I was able to replicate results so i know its useful to see how various settings affect the sim.

 

I dont think there is any magic fix to this.. its just a shortcoming of fsx and how it handles the high res cloud pixels with sparse grid.. pretty poorly as you can see in this situation anyway, except maybe shrinking screen sizes or using Oculus Rift

I have P3D v2.0, my goal is to try this same scenario in that soon.  

 

Here were my latest examples:

 

17.96 fps 4SGSS 4xSSAA 

32.50 fps 8xSUPERSample 4xSS AA (flickery, esp in other cities like NYC, particularly distant buildings and airplanes on ground, NYC is the worst if you try New York X for instance, distant land is jagged too)

20.83 fps 2SGSS 8SQ AA, fsxcfg AA at 4,8

20.71 fps 2SGSS Appcontrol AA, fsxcfg 4, 8 (i think it was)

 

Keep in mind other areas that aren't as "busy" as a KSEA or NYC generally get 24-35 fps even with some clouds and weather.  For example, if you run the fsmark07 test with my settings and use 4xSGSS, i'm hitting around 31 fps.

 

Here is one test scenario NI screenshot (2sgss):

 

FSX.cfg relevant settings (AA below varied depending on what i set dx10 fixer to):

 

Mode=5760x1080x32

Anisotropic=1

AntiAlias=1

 

[TERRAIN]

LOD_RADIUS=6.000000

AUTOGEN_DENSITY=2

BATCH_AUTOGEN2=1

DETAIL_TEXTURE=1

MESH_COMPLEXITY=90

MESH_RESOLUTION=23

//SWAP_WAIT_TIMEOUT=2  

TERRAIN_MAX_AUTOGEN_BUILDINGS_PER_CELL=800  //800 for very dense and up  //1500 another 

TERRAIN_MAX_AUTOGEN_TREES_PER_CELL=1110    //1100 for very dense and up  //1900 another

TEXTURE_RESOLUTION=29

WATER_EFFECTS=4

//IMAGE_PIXELS_FOR_AUTOGEN_POLYGONS=2048  //new

SWAP_WAIT_TIMEOUT=10

 

[GRAPHICS]

TEXTURE_MAX_LOAD=2048

AC_SELF_SHADOW=0

AIRCRAFT_REFLECTIONS=1

AIRCRAFT_SHADOWS=1

ALLOW_SHADER_30=1

COCKPIT_HIGH_LOD=1

D3D10=1

EFFECTS_QUALITY=2

ForceWindowedVSync=0

//ForceFullScreenVSync=1 //if driver set to app manage

GROUND_SHADOWS=0

HIGHMEMFIX=1 

IMAGE_QUALITY=0

LANDING_LIGHTS=1

MultiSamplesPerPixel=4  

MultiSampleQuality=8

NUM_LIGHTS=8

See_Self=1

TEXTURE_QUALITY=3

Text_Scroll=1

SHADER_CACHE_PRIMED_10=1693500672

SHADER_CACHE_PRIMED=1693500672

ForceFullScreenVSync=1

 

[bUFFERPOOLS]

Poolsize=0 

Rejectthreshold=98304  //i dont think this is needed with 0

 

Other settings:

FIBER_FRAME_TIME_FRACTION=0.15 

bloom off

TEXTURE_BANDWIDTH_MULT=120

TextureMaxLoad=9

TransitionTime=4.000000

UPPER_FRAMERATE_LIMIT=36

WideViewAspect=True

 

 

[Weather]

AdjustForMagVarInDialog=1

CLOUD_COVERAGE_DENSITY=8

CLOUD_DRAW_DISTANCE=3

DETAILED_CLOUDS=1

DisableTurbulence=0

DownloadWindsAloft=0

DynamicWeather=2

MaxGustRampSpeed=200

MaxGustTime=500

MaxVarRampSpeed=75

MaxVarTime=50

MinGustRampSpeed=1

MinGustTime=10

MinVarRampSpeed=10

MinVarTime=5

THERMAL_VISUALS=0

TurbulenceScale=1.000000

WeatherGraphDataInDialog=0

WeatherServerAddress=fs2k.zone.com

WeatherServerPort=80

WindshieldPrecipitationEffects=1

 

[TrafficManager]

AIRPORT_SCENERY_DENSITY=1

AirlineDensity=0

FreewayDensity=7

GADensity=0

IFROnly=0

LeisureBoatsDensity=0

ShipsAndFerriesDensity=0

 

 

[Trans_27 Pilot Chase 59]

And people still say FSX is the future... If a card like this can't handle more than 11FPS, no card ever will!!! This thing is a beast.

 

 

It's not the card as this point in the game.

 

It's FSX's old code and utilizing only one core. If it would run across all 4 cores, we'd be happier than we've ever been. 

[theskyisthelimit 48]

I've done similar testing and tweaking (so far) with P3D v2.0.. found in this thread in case anyone is interesting in reading.

[gboz 157]

 

 

Here is one test scenario NI screenshot (2sgss):

 

Hi again.

 

Really admire your continued search for the FSX holy grail.  .

 

Are these NI settings and the rest related to DX9 or DX10?

 

gb.

[theskyisthelimit 48]

Hi again.

 

Really admire your continued search for the FSX holy grail.  .

 

Are these NI settings and the rest related to DX9 or DX10?

 

gb.

 

Thanks  Though i've begun full migration to P3D which gives better performance for me at least.. those screen shots are all pertaining to DX10 with FSX.. i have stuck with dx10 due to the better performance.

 

Currently this is what i have for FSX, though i get a bit of a boost just using 4xsupersample instead, but in most cases 4sgss is ok:

 

[gboz 157]

Thanks  Though i've begun full migration to P3D which gives better performance for me at least.. those screen shots are all pertaining to DX10 with FSX.. i have stuck with dx10 due to the better performance.

 

Currently this is what i have for FSX, though i get a bit of a boost just using 4xsupersample instead, but in most cases 4sgss is ok:

 

FSXfinal4sgss.png

I thought it was pretty well established that the only antialiasing setting that works with DX10 in NI was transparency supersampling and the rest has to be done through the fsx.cfg .

 

Doesn't that make your 4X NI setting redundant.

Also setting "Override app settings" could be turning off fsx AA settings. ??

 

gb.

[theskyisthelimit 48]

I thought it was pretty well established that the only antialiasing setting that works with DX10 in NI was transparency supersampling and the rest has to be done through the fsx.cfg .

 

Doesn't that make your 4X NI setting redundant.

Also setting "Override app settings" could be turning off fsx AA settings. ??

 

gb.

 

Yeah your right.. i had been tinkering with some settings just to see their effect if any and thats where i last had it when i was using fsx..

 

It should be more like this:

 

I believe the following holds true if i'm not mistaken....

It should be app controlled, because NI cant override anything with DX10 as far as AA goes (and fsx).

 

The Nvidia Pre-defined FXAA usage, if set to allowed, would make less jaggies on the spinning aircraft on the start screen and add some "after the fact" AA.

 

Incidentally, the -1.00 LOD bios helps with blurring that is the result of using SGSS.. i guess if using supersample (as i do if heavy weather/clouds), then maybe this line should be defaulted or different.

 

Texture filtering Negative LOD bias is set to "Allow" so that the -1.00 setting can actually work.

[gboz 157]

Yeah your right.. i had been tinkering with some settings just to see their effect if any and thats where i last had it when i was using fsx..

 

It should be more like this:

nifsx.JPG

 

I believe the following holds true if i'm not mistaken....

It should be app controlled, because NI cant override anything with DX10 as far as AA goes (and fsx).

 

The Nvidia Pre-defined FXAA usage, if set to allowed, would make less jaggies on the spinning aircraft on the start screen and add some "after the fact" AA.

 

Incidentally, the -1.00 LOD bios helps with blurring that is the result of using SGSS.. i guess if using supersample (as i do if heavy weather/clouds), then maybe this line should be defaulted or different.

 

Texture filtering Negative LOD bias is set to "Allow" so that the -1.00 setting can actually work.

I also see that you have set the frame rate limiter to 35 in NI.

Shouldn't that be 36 for your 144Hz monitor?

With 35 set standard Vsync will try to give you 18 FPS. ??

 

gb.

[theskyisthelimit 48]

I also see that you have set the frame rate limiter to 35 in NI.

Shouldn't that be 36 for your 144Hz monitor?

With 35 set standard Vsync will try to give you 18 FPS. ??

 

gb.

 

I have it set to 1/4 refresh rate.. it was the closest i could get to 36 fps in NI using external limiter.. seems smooth though.. there isnt an option for 36 in NI.. though admittedly i keep flipping back and forth at times.. one time internal 36 with NI 1/4.. another external lock.. granted i think i only gain 1.5 fps by using external.