Thanks Paul for summarizing.

I would also remove the vsync in windowed. It's only a troublemaker.

TextureMaxLoad=30 never saw the point of this as of late.

TriLinear=1 <--- Good - why not Anisotropic?

Rest I agree.

 

What are you Nvidia Inspector settings? Please post some pics...

Here are my settings, excuse my ignorance, kow do I "remove vsync in windowed " ?

 

 

Thanks Paul for summarizing.

I would also remove the vsync in windowed. It's only a troublemaker.

TextureMaxLoad=30 never saw the point of this as of late.

TriLinear=1 <--- Good - why not Anisotropic?

Rest I agree.

 

What are you Nvidia Inspector settings? Please post some pics...

 

Yeah, I take out Vsync, too, as it's a frame hog - and besides, the latest versions of Inspector have it selectable and it works.

Aniso and Inspector settings... I use Tabs's (Ryan M - PMDG) suggestions, saved as profiles, and then load each as I see fit, plus I might change the odd thing here and there, Word Not Allowed. We were talking about limiting internally, while not limiting the OS a few days ago - it was your suggestion, and I think you were right, inasmuch as it inhibits the OS - the CPU - and so should be unlimited instead. I tried this and I'm inclined to agree, so now it's set that way.

There's nothing magic that you are not aware of. I'm pretty conservative generally, as it can certainly have an impact in heavy areas, such as the VancouverPlus, Seattle or New York sceneries.

 

The major settings that have had the most impact for me, are moving the upper limit to 27, TBM to87. The [Display]TML back to 3, {Graphics] TML to 2048, (all calculated) and reducing the FFTF to .1 This was your suggestion, too - I originally worked with this one in the early Bojote days, with the C2D and the i7-950, and arrived at .15 as the lowest setting. It can go lower, now with the higher performance gpu.)

 

@ Paul:

It's in the [Graphics] section of your fsx.cfg, Paul:

 

Just delete the two lines:-

 

ForceFullScreenVSync=1

ForceWindowedVSync=1

 

@ Word Not Allowed: (Sorry, I missed this piece of your post!)

 

TextureMaxLoad=30 never saw the point of this as of late.

 

This is a multiplyer, Word Not Allowed, and has a huge impact on the maximum texture data that can be used by the gpu.

 

Global Texture Resolution * (Maximum Textures Per Frame * Texture Bandwidth Multiplier) / Target Frame Rate = Maximum Bytes Per Frame

This is a multiplyer, Word Not Allowed, and has a huge impact on the maximum texture data that can be used by the gpu.

 

Global Texture Resolution * (Maximum Textures Per Frame * Texture Bandwidth Multiplier) / Target Frame Rate = Maximum Bytes Per Frame

 

Nono, I know what it's for. I only ment do you really see an impact from it? Or is it there just for the sake of thinking it does anything?

Weeeeeell....I don't really remember, Word Not Allowed, as it's all slowly come together over the last two/three years or so. As we're now finding - the newer hardware is making some of the Bojote tweaks obsolescent. I know that he preached a "30" setting for the Texture Bandwidth Multiplier (provided your frames were set to unlimited), but I've never tested against that, as it's too unstable. That's fine for some, but unlimited has it's own set of problems, and I think that a constant limit, which will take care of the varying graphical and cpu loads, is a better long-term solution. Nowadays I'm working the gpu's settings via mathematics, rather than by guess and by golly, and the machine is working much better than it was three months ago - largely through monitored and documented test flights - though I haven't done any with unlimited frames. Similarly with the poolsize - royally screwed up now with the increased used of the (fairly) new SSD's on the market.

Global Texture Resolution * (Maximum Textures Per Frame * Texture Bandwidth Multiplier) / Target Frame Rate = Maximum Bytes Per Frame

 

Is this the only math formulla?

Weeeeeell....I don't really remember, Word Not Allowed, as it's all slowly come together over the last two/three years or so. As we're now finding - the newer hardware is making some of the Bojote tweaks obsolescent. I know that he preached a "30" setting for the Texture Bandwidth Multiplier (provided your frames were set to unlimited), but I've never tested against that, as it's too unstable. That's fine for some, but unlimited has it's own set of problems, and I think that a constant limit, which will take care of the varying graphical and cpu loads, is a better long-term solution. Nowadays I'm working the gpu's settings via mathematics, rather than by guess and by golly, and the machine is working much better than it was three months ago - largely through monitored and documented test flights - though I haven't done any with unlimited frames. Similarly with the poolsize - royally screwed up now with the increased used of the (fairly) new SSD's on the market.

 

OK Paul, let's see.

If you are saying that the math is working so well for you, elaborate please. What do you mean GPU's settings via mathematics?

No, Johan. First I have to say this is not my origination: I've gleaned much of this with a smidgeon of knowledge and a few posts (hopefully refined it a little) around the web -particularly on the SimOutHouse site.

 

The formula quoted above verifies that the theoretical numbers for TML, TBM, chosen frame limit and maximum data per frame all agree with earlier calculations that derive those numbers.

 

The first formula finds the maximum # of bytes per frame, and it depends upon your gpu: for Nvidia gpu's you need Inspector - a great tool: for ATI - I haven't a clue.... sorry... Inspector gives you some numbers, thus.

 

Memory Clock in MHz.... in the case of the EVGA GTX 580, it's 2005MHz

Given - GGDR3 is DDR2 and uses a multiplier of 16, GDDR5 is DDR3 and uses 24

Target Frame Rate = whatever you think fits..

Maximum bytes per frame - this is the result we're looking for.

 

To find the Maximum bytes per frame:-

 

(GPU Memory Clock * GDDR multiplier) / Target Frame Rate = maximum bytes per frame

 

So -

 

(2004 * 24) / 30 =

 

48096 / 30 = 1603.2

 

MBF = 1602.3

 

To find the TEXTURE_BANDWIDTH_MULTIPLE:- (again - the GTX 580)

 

(Maximum Bytes Per Frame / GLOBAL_TEXTURE_RESOLUTION) * 100 = Texture Bandwidth Multiplier

 

(1602.3 / 1024) * 100 = TBM

 

1.5656 *100 = 156.56

 

TBM = 156.56

 

**********************************************************************************************************

 

To verify your figures:-

 

TEXTURE_MAX_LOAD * (TextureMaxLoad * TEXTURE_BANDWIDTH_MULTIPLIER ) / Target Frame Rate = Maximum Bytes Per Frame

 

102.4MB * ( 3 * 56.56) / 30 =

 

102.4 * 469.68 /30 =

 

48095.232/30 = 1603.2

 

**********************************************************************************************************

 

Play with these sizes, and you will quickly see by testing just how close, or correct - these figures are, and how much impact a 'very wrong' figure can change the output.

 

The guys at the DigitalThemePark (don't know who - sorry) created a spreadsheet with a collection of examples using different gpu's, but without any underlying explanation.

 

https://docs.google....1aVhoNXc#gid=10

So if my TEXTURE_MAX_LOAD is 4096, but my only 4096 textures are the NGX ones, I still should use the 4096?

 

TEXTURE_MAX_LOAD * (TextureMaxLoad * TEXTURE_BANDWIDTH_MULTIPLIER ) / Target Frame Rate = Maximum Bytes Per Frame

 

102.4MB * ( 3 * 56.56) / 30 =

 

And why do you say 1024 is 102.4MB? I always believed that was the size of the texture: 1024x1024 or 4096x4096.

...reckon so, Word Not Allowed, otherwise (I believe) it'll get rolled back to the 1024 default.

 

Y'know - you're forcing me to do a lot of work.... I guess this will be in the SDK. I shall go look... Back in a while, unless you find it first. It depends upon which formula you're working with, whether it's using just multiple, bytes, k or M.

To verify your figures:-

 

This last formula is nothing else than a reversed upper formula, with one 3 multi, one 30 divider, and one decimal place less. I know math that much!

 

...reckon so, Word Not Allowed, otherwise (I believe) it'll get rolled back to the 1024 default.

 

You misunderstood my question possibly.

If I were to use your formula from above, and my fsx.cfg setting is 4096, but I am not using 4096 textures, should I still input 4096 into the formula, or should I input 1024?

 

Y'know - you're forcing me to do a lot of work....

 

Not forcing you to do anything. I just expect some explanations if you come with such claims. That is a pretty big thing you just pointed out here, so it would be useful if everything would be out open in the clear.

Besides if it can be proven that it's working, I will put this into my FAQ I'm currently assembling.

 

https://docs.google....1aVhoNXc#gid=10

 

Oh sorry! I totally missed that one. But still would like to know where these formulas are coming from...

 

I'll be checking this tonight.

You have only to experiment with the figures, Word Not Allowed. I first heard about this some time ago, then saw the spreadsheet, then wondered how it was calculated, so I did some more digging, constructed my own spreadsheet, which allows variation of the TextureMaxLoad parameter, found two other forums where one, at least seemed to understand it (one was quoting RAM bandwidth) and so I changed the odd parameter names to say what they meant technically (Bandwidth does not mean Frequency), downloaded all the tech specs on the 560Ti and the 580, and then did a week's worth of flying and documenting.

So what is your conclusion, what is better since you applied the calculated tweaks?

Can you point me to the original thread?

 

Another thing putting me off a bit is that they claim there is a performance difference, and not a dismissive one at that, between texture quality set at high performance and high quality. 80 to 93 fps? What the hell did they test!?

I'm now using 2048, as a TMLOAD, a limit of 27 and a TBM of 87, with the tml still at 3. It changes the way I tweak, Word Not Allowed: It's very smoooooth with these settings. I've been up and down the limit, which changes the TML, I've been up and down the tml - and one can come up with some pretty high data loads with - as quoted earlier - going from 3 to 30, without any other change. These formulae are a good guide as to what I've changed, and can be attached to a particular flight. As I had said earlier - the test flight is over VancouverPlus, with OrbX PNW, Grovestine's Victoria 2010, water a 6, REX 2048 cloud textures - Building Storms selected - ~30 GA and AI, LOD=4.5,MTX, road, ship, boat and airport traffic up some - into the low teens, EZCA, Shade, TIR, BP=0, "AccuFeel", TTFT=.1, not to forget - running three monitors.. It's pretty heavy, yet she now sits at a very constant 26.9 frames, and is as smooth as silk. It is way better than it was two weeks ago.

 

Yeah, gimme a sec I have to dig..