fsx-se A simple VAS usage analysis

[Bjoern 1,157]

I've run into VAS consumption issues lately and was curious as to what could be the cause, so I ran some tests and scribbled down some numbers.
This post is not a comprehensive test to determine the impact of addon x or y on your VAS consumption, but rather a helper as to what type of content and settings to look at when you run into issues. It kind of supplements the 324534218734287342 other analyses already done by others (or at least Word Not Allowed's, back in the day).

All numbers and observations are for my FSX configuration. Your mileage may very well vary.
 

Computing environment:

• i5 4670K at 4,0 GHz (in turbo mode)
• 8 GB 2400 MHz RAM
• GTX570
• Sandisk 128GB SSD (Windows) & Crucial 256GB SSD (FSX)
• Windows 10 x64
• FSX Steam

All updated and featuring the latest drivers.


FSX environment:

• Ultimate Terrain X USA v1.6
• 410 AI airlines plus a global, autogenerated GA AI file (136MB) with mostly FSX native models, i.e. tons of AI
• Water, cloud, etc...textures with stock resolution and filesize
• FreeMeshX

 

 

FSX settings:

• 1680x1050 and fairly high, but pretty much irrelevant in the context of these tests as configuration changes and the impact on the result attained with my baseline configuration will be mentioned for each step anyway

 

The readout for the remaining VAS was obtained via FSUIPC. Take note that lower numbers are worse than higher numbers.
Settings used in my FSX are not
 

 
The first test was observing the impact of aircraft, AI, mip mapped textures for AI aircraft, terrain mesh and various fsx.cfg settings.
 
For this test, I've put the default C208, at 6:30am on this very Friday (Aug. 26th) on the active runway at KMSP. MSP uses a custom airport layout file, retrograding it into its mid to late 1980s state. Weather was the default "cold fronts" theme, viewpoint was exclusively from the VC and FSX was restarted after every configuration change, even if it was not really necessary (e.g. after changing aircraft).
The C208 was used to establish a baseline result in the simulator to detemine the relative VAS impact.
Configuration changes were done individually from the baseline config.
 
I didn't care to write down the exact radout in bytes, so the numbers you see below are rounded.

 

VAS remaining in the FSX main menu:

• 3.55 GB

VAS remaining in the baseline in-simulator scenario:

• 2.68 GB

VAS remaining with the (modded) RAZBAM Metro 3 as the aircraft (one of the two OOM offenders):

• 2.13 GB (-0.54 MB from baseline)

VAS remaining with the (modded) RAZBAM Metro 3 and deactivated airport layout files as the aircraft (ADEX/AFX/AFCAD):

• 2.14 GB (-0.54 MB from baseline; +10 MB gain compared to the previous setup)

VAS remaining with the (modded) Sky Simulations DC-9 as the aircraft (the second OOM offender):

• 2.50 GB (-0.18 GB from baseline)

VAS remaining with Tom Ruth's (modded) 727-200ADV as the aircraft (no OOM offender):

• 2.50 GB (-0.18 GB from baseline)

VAS remaining with the LOD radius increased from 4.5 to 5.5 in the FSX.cfg:

• 2.54 GB (-0.14 GB from baseline)

VAS remaining with FreeMeshX disabled:

• 2.70 GB (+0.02 GB from baseline)

VAS remaining with AI disabled (flight plan folder deactivated in the scenery.cfg and path to aircraft folder deactivated in the fsx.cfg; affects the visible portion of 227 aircraft in the AI bubble*):

• 2.80 GB (+0.12 GB from baseline)

VAS remaining with AI textures without mip maps (reducing a 1024 px DXT5 DDS texture from 1.33 to 1 MB; affects the visible portion of 227 aircraft in the AI bubble):

• 2.72 GB (+0.04 GB from baseline, VAS hit for AI in the scenario: -33%)

VAS remaining with AI textures without mip maps and ground vehicles (road and all ship traffic) at 0% (Note: I run custom road traffic):

• 2.74 GB (+0.06 GB from baseline, +0.02 GB gain from the previous setup)

 

 

Observations:

• Airplanes are a huge VAS offender, especially ones with a lot of high resolution textures like the RAZBAM Metro.
• The LOD radius setting in the fsx.cfg will also incur a hit on remaining VAS
• Disabling AI helps if there is a lot of it
• Removing mip maps from AI textures is a way to save a bit of VAS space.

 

 

 

For the second test, I've put the default C208 onto the active runway at JFK. The date was set to this Friday (Aug. 26th) at the default "Day" time of day (around noon). The weather was again set to "cold fronts". The viewpoint was the VC and FSX was restarted after every configuration change. AI and custom airport layout were disabled for the test. FreeMeshX and Ultimate Terrain X remained enabled. The scenerio was lanuched and allowed to settle for a bit before noting the remaining VAS.
 
The objective for this test was observing VAS usage in relation to the scenery settings picked from the FSX "display" menu. Configuration changes were made sequentially instead of individually. The interesting part of the results is the relative decrease in remaining VAS as one visual feature after another is enabled and set from its minimum to its maximum.

 

VAS remaining in the FSX main menu:

• 3.55 GB

VAS remaining in the baseline in-simulator scenario (with all scenery sliders left and off):

• 3.27 GB

VAS remaining after putting the "Scenery Complexity" slider from "Very Sparse" to "Extremely Dense":

• 3.12 GB (-0.15 GB from baseline)

VAS remaining after putting the "Autogen Density" slider from "None" to "Extremely Dense":

• 3.00 GB (-0.12 GB from previous configuration; note that I use AutogenDescriptions_Min.spb)

VAS remaining after putting the "Level of Detail Radius" slider from "Small" to "Large":

• 2.74 GB (-0.26 GB from the previous configuration)

VAS remaining after putting the "Mesh Complexity" slider from "0" to "100":

• 2.73 GB (-0.01 GB to the previous configuration)

VAS remaining after putting the "Mesh Resolution" slider from "305 m" to "1 m":

• 2.65 GB (-0.08 GB to the previous configuration)

VAS remaining after putting the "Texture Resolution" slider from "10 m" to "7 cm":

• 2.55 GB (-0.10 GB to the previous configuration)

VAS remaining after enabling "Land Detail Textures":

• 2.53 GB (-0.02 GB to the previous configuration)

VAS remaining after putting the "Water Effects" slider from "None" to "Max 2.x":

• 2.45 GB (-0.08 GB to the previous configuration)

VAS remaining after putting the "Special Effects Detail" slider from "Low" to "High":

• 2.45 GB (no change to the previous configuration)

 

Quick follow-on testing using the previous configuration above as a starting point:
 
VAS remaining after putting the "Level of Detail Radius" slider from "Large" to "Small":

• 2.88 GB (-0.39 GB from baseline; +0.43 GB from the previous configuration)

VAS remaining with AI and custom airport layouts enabled (with LOD radius at "Large"):

• 2.29 GB (-0.98 GB from baseline and -0.16 GB from the second to last configuration with 479 aircraft in the AI bubble*)

 

 

Observations:

• The worst offender in terms of VAS usage is, again, the LOD detail radius, followed by scenery complexity and autogen
• Other scenery detail options may be adjusted to conserve VAS. Putting the maximal mesh and texture resolution at 19 m and 60 cm respectively and dropping water to 2.x low conserves about 0.11 GB of VAS.

 

 

 

A quick third test dealt with the "DisablePreload" line for the fsx.cfg. The objective was investigating whether the presence of the line in the fsx.cfg made any difference in VAS consumption by preloading the location of your default flight (BGR in my case) when you wish to start from another airport. This builds upon the scenerio for the first test (i.e. C208 at MSP at dawn) with the difference of using non-mip mapped textures for the AI aircraft.
After launching FSX, I let it sit in the main menu for five minutes and watched FSUIPC's reported remaining VAS before loading the testing scenario.
 
VAS remaining in the FSX main menu without "DisablePreload=1" in the fsx.cfg :

• 3.56 GB

VAS remaining after starting the testing scenario without "DisablePreload=1" in the fsx.cfg :

• 2.72 GB

VAS remaining in the FSX main menu with "DisablePreload=1" in the fsx.cfg :

• 3.56 GB (no change compared to previous config)

VAS remaining after starting the testing scenario with "DisablePreload=1" in the fsx.cfg :

• 2.72 GB (no change compared to previous config)

 

 

Observation:
The "DisablePreload" line in the fsx.cfg has no practical effect in FSXSE. Whether this is due to Dovetail's improved scenery unloading or a broken basic functionality due to recompiling the simulator, I can't say. Maybe the unloading only affects full fledged add-on sceneries like airports.
 
 
 
 
Bonus tests:
Impact of texture-related settings on FSX' VAS. The setup is as in test #2, with the difference being that AI and airport layouts are enabled from the start.
The starting point was "Global Texture Resolution" at "Very Low" and all eyecandy settings on the aircraft page disabled. Settings were again enabled sequentially.
 
VAS remaining in the baseline in-simulator scenario:

• 2.40 GB

VAS remaining with all aircraft eyecandy settings enabled :

• 2.40 GB (no change to baseline configuration)

VAS remaining with "Global Texture Resolution" at "Maximum" (4096 px in FSXSE) :

• 2.30 GB (-0.10 GB to baseline or previous configuration)

VAS remaining in DX10 mode:

• 2.35 GB (+0.05 GB to previous configuration; note: No anti-aliasing or anisotropic filtering are set since I normally don't use DX10)

VAS remaining in DX9 mode without anti-aliasing and anisotropic filtering (forced via NVidia Inspector):

• 2.30 GB (no change to previous DX9 configuration)

VAS remaining in DX9 mode with anti-aliasing and anisotropic filtering enabled in FSX:

• 2.25 GB (-0.05 GB to previous DX9 configuration)

 

 

Observations:

• The global texture resolution seems to have an impact on VAS
• The AA/AF impact from the FSX-internal options as well as the VAS improvement in DX10 mode is a bit odd and could have another cause (AI aircraft becoming visible or else)

 

 

 

Conclusions:
It's hard to pinpoint a single cause for OOM problems, but testing on my configuration shows that the user aircraft, AI and level of detail radius are the worst offenders in terms of VAS use. I'm a bit disappointed that there's no other, less important setting that can be adjusted since I'm very much fond of a living 3D world with a reasonable area of detailed ground textures. Furthermore, these tests were static. The benefits or drawbacks of each setting on an entire (long) flight were therefor not determined. One could, however, say that any VAS usage improvement in a static scenario (or at the beginning of a flight) may prove useful during the flight itself. Also note that the impact of AI models themselves is fairly low, accounting for just 0.16 GB even in the NYC area with its three busy, major airports.
 
Testing also shows that the general rule of thumb for FSX remains:
Anything that needs to get loaded from disk negatively affects the amount of FSX' remaining VAS.

 
Hope this is useful to someone. Just wanted to avoid throwing the paper with my notes away.
 
 
 
 
* "AI bubble" describes the radius in which FSX creates and tracks AI aircraft. AI only incurs a VAS hit from the moment its lowest detail model becomes visible on screen, making it hard to determine how many AI aircraft are actually visible around the user aircraft.

[AviatorMoser 353]

Cool test! Well understood!

[stans 702]

Everyone wants and developers provide 4096x4096 textures for scenery and aircraft, then when FSX crashes due to an OOM they wonder why.  FSX is a 32 bit program.  You can have all the system RAM in the world, but FSX can use no more than 4 GB.  FSX is notoriously poor at managing its VAS and when you dump huge texture files into the mix, then add tons of AI, the VAS is quickly consumed.  I understand that the Steam Edition of FSX is actually much better with its VAS management, but if you add in enough huge texture files, enough AI, and set high graphics settings it will OOM as well.

[Bjoern 1,157]

Just another small follow-up. I've played with DX10 mode (and the free fixes) and found that it indeed helps with VAS usage. Recreating the Metroliner flight that made FSX run out of memory before, I've ended up with almost 0.9 GB of VAS remaining after shutting down instead of 0.12 GB (and a subsequent simulator shut down). Mind you, this was with LOD radius at 5.5, worse weather and more AI traffic (since I modified the GA portion quite a bit last night). So the benefit of DX10 mode can only be measured over the course of a flight instead of a static situation.

 

Seriously considering buying the payware fixer now as long as it reduces the FPS hit in clouds and fixes light (nav, landing, etc...) effects.

[Petraeus 147]

Thanks for this thorough and meticulous study. Lots of hard work, but well worth it.

 

Personally I always go for a lower LOD Setting, and don't actually see any difference in the scenery - maybe my weak eyes!

[Alan_A 1,650]

This is enormously helpful.  I usually don't fly in VAS monsters, or fly in VAS-eating scenery.  But I'm weighing whether to get in on the FSL A320, which means I'm wondering how best to tweak settings and still wind up with some sort of living 3D environment, as opposed to a flat, post-apocalyptic wasteland with an airliner flying through it.  Will run some experiments based on your findings.  Many thanks!

[Bjoern 1,157]

You're very welcome, guys.

 

 

After the successful Metro test, I've reflown the second flight that gave me an OOM with DX10 enabled. Purchased and installed the DX10 Fixer beforehand to fully tap the potential of DX10. Took my time to get out of MSP in the DC-9 during the late afternoon, flew over STL and BNA which both add a bit of overhead in terms of AI traffic and finally landed without hitches in ATL in the evening. Remaining VAS dipped as low as 300 MB on approach, but save for some small weather injection and scenery model loading hiccups and slowdowns in bloom-intensive areas (airports), FSX never really complained. And this is with a LOD radius of 5.5 and transition from day to night textures on the ground. The shaders provided by the Fixer even greatly reduced the framerate impact of clouds, not to mention the additional eye candy! Impressive! And still only half as expensive as a P3D license...

[jwyman 0]

I know this is somewhat of an old thread but this is fantastic info!

I have recently built a super-charged machine and am getting OOMs after using high detailed A/C and scenery. I have found out that Ultimate Traffic Live is injecting about 1 GB into the VAS arena.

I am going to disable that and re-fly my flight from KMIA to KLGA using teh Aerosoft A321. This is the configuration that gave me the OOM....

 

 

Jim

[Bjoern 1,157]

Thanks!