Hi all,

 

I've still one problem that's holding me from enable dx10 for good.

 

I've done some tests which shows that the performance with dx10 in clouds is really bad.

 

Some results: Same situation default cessna EHAM 1500ft

 

Both rex clouds 512x512 dxt5 (just for this test)

 

Both 3 layers created cumulus 6/8, 7/8, 5/8 with 10nm vis.

 

dx9

AA is 8xS and 2xSGSS

gpu load 74%

constant 30 fps

730mb vram

 

dx10

2x SGSS which is even worse AA quality in relation to dx9. 4x SGSS equals 8xS 2x SGSS dx9

gpu load 97 - 99% (stutter)

15 - 20 fps

738mb vram

 

Does someone has some contact with the guy who made this fantastic shader fixes?

I really like to solve this. Then is almost everything right. At least it is flyable with my home weather sitation

 

I really like dx10 because Aerosoft EHAM 737 ngx is normally around 24 fps in dx9 and with dx10 it is constant 30 fps and everything is also a lot sharper and smoother.

 

Kind regards.

I personally do not have the problems you are having but then I have my display drivers set to the default for FSX in NI and the display drivers. I just have AA/AF controlled by the application. In any case, you can contact Stevejp by looking at the pinned topic above entitled DX10 Installation Problems. He has the address to his blog in his signature. Hope this helps.

 

Best regards,

Jim

Ive just set NI to the defaults for fsx so AA/AF is controlled by FSX itself.

 

then the results are:

 

dx10

AA / AF inside fsx

gpu load 68%

constant 30 fps

730mb vram

 

I will try to contact him.

Hi

 

I will look at the clouder shader, busy tweaking vc shadowing in jets at the moment. I don't have an nvidia card and I find their terminology hard to follow, so i stuggle with what the settings mean!

 

Sgaa is ssaa I think, i take it you dont have MultiSamplesPerPixel set for dx10

 

I would actually start by testing performance 4x msaa in both so no Sgaa, then try 4x samples with Sgaa enabled as that's easy to set in both, before we look at 8x.

 

Is 8xS 8 samples or 4samples with two cheap coverage samples?

 

8xS is CSAA which in in AMD terms is called 4x EQ - so its four full samples plus some coverage samples.

 

Nvidia call its A- whilst AMD call it B+

 

Default in DX10 is 4x (AMD) 4xQ (Nvidia) so 4 four full samples

 

I woudl start therefore by comparing default DX10 to 4xQ in DX9, then add SGAA into the equation.

 

I agree by the way that there is something going on in DX10 with clouds - they do seem much slower

Thanks for your time.

 

I dont know much about the difference how AMD and Nvidia handle AA.

 

But I will start comparing with the options you suggested.

 

But I also know that whatever AA setting I set in dx10 that eliminates a little bit of the shimmers, the performance in clouds already drops. So indeed there is something going on in dx10 with clouds.

 

Is this the same with high quality AA AMD?

Sorry - I confused myself.

 

I created 4 layers of clouds and then viewed them in DX10 and DX9. I had twice the fps in DX10 at 8x SSAA

which is equivalent to 8xEQ +SGAA in Nvidia against the same in DX9. I had 6 fps with DX9 and 13 with DX10!

 

Despite that I thought that there was a DX10 issue because when i saved the frame and viewed it with a diagnostics tool it seemed to be taking half the time in the GPU for DX9. Eventually I realised that the DX9 frame didnt include the forced driver settings and so I was measuring 2x AA for DX9 within that tool. The AA and IQ had been the same back in FSX at the framerates above.

 

So my conclusion is that it cloud drawing looks fine in DX10 so far as I can see.

 

The most important thing for you to check is that you are comparing apples with apples.

 

If you perhaps have MultiSamplesPerPixel set to 8 - i ask because Paul suggests setting it in his doc; then in the first test you describe, DX10 would have been running at a cripplingly high level 8xQ + SSGA which is much harder for the GPU than 8xS +SGAA which is really 4xQ + some cheaper coverages samples + SGAA.

 

So the first test is to get both in 4xQ without SGAA and measure, then add the same SGAA setting to both and compare again. This will tell us if DX10 is exhibiting an issue with layered clouds.

Ive done some more tests in the meanwhile, and I think you're indeed right.

 

I didnt really compared apples with apples but what I did, is set the AA so, that the IQ is almost the same for dx9 and dx10.

 

Which means that dx10 needs a higher AA level for the same IQ in dx9 (almost no shimmers)

 

I can set the transparency supersampling the same for both dx9 and dx10 which is 2x sparse grid supersampling.

 

But I cant set the normal Antialiasing the same. So it wouldnt be a completely fair comparision.

 

In dx9 I set it to 8xS which is 1x2 SS + 4 MS.

In dx10 this setting is ignored so therefore AntiAlias=1 and MultiSamplesPerPixel set to 8.

 

That's what I will try. And see what for performance it gives

 

What I saw is that dx9 with 8xS and 2 x Sparse grid is the same as FSX AA with multisamples = 8 in the cfg with 4x Sparse grid.

 

Sparse grid supersampling is known for that is causes bad fps in cloudy skies. So that's I think why I have half the frame rate I get with dx10 then I would have in dx9.

 

I will post my results later.

I've done some more tests.

 

There must be a problem how fsx handles nvidia's AA in dx10 with clouds.

 

The same situation just on a default airport with stormy weather profile.

 

dx9 constant 30fps with 70-80 gpu load.

dx10 20-25 fps with 97-99 % gpu load with stutters.

 

In dx9 I set it to 8xS which is 1x2 SS + 4 MS with 2x sparse grid supersampling.

In dx10 I set AntiAlias=1 and MultiSamplesPerPixel set to 8 with 2x sparse grid supersampling.

 

Looks to me that my antialias setting with dx9 are harder but my performance is better than with dx10.

 

I hope that there is some way we can fix how dx10 handles AA. Because I really need 4x SGSS to remove almost all shimmering. And that's even harder on the system than 2x SGSS.

 

But without clouds and AA set to 4x SGSS my fsx looks wonderfull. It has never looked so good before.

 

Too sad that the weather overhere contains a lot of clouds.

I stil think you are letting yourself be confused by the nvidia marketing dept.

 

http://www.nvidia.com/object/coverage-sampled-aa.html

 

 

In nvidia speak 8x is actually a dx10 sample setting of 4 and a quality setting of 8

 

To get this in fsx you need to set

MultiSamplesPerPixel=4

MultiSampleQuality=8

 

So I would set the samples and quality that way in fsx.cfg and then try and add super sampling in nvi

 

In nvidia speak MultiSamplesPerPixel=8 is 8xQ the next level up and it's a big step in processing

 

 

It looks like my problem is solved ^_^ But I will do some more testing tommorow.

 

Im reading the nvidia page you posted. I don't know much about AA techniques but only what I need in fsx so there isnt any shimmer anymore.

 

But ive forgot that indeed dx10 handles AA settings different then dx9.

 

I've set my AA in nvi to 4x SGSS again which is the hardest AA mode for nvidia (I think) But it is necessary to remove most of the shimmers and jaggies.

 

4x SGSS combined to your settings: MultiSamplesPerPixel=4 MultiSampleQuality=8

 

My gpu load went from 99% with stutters to 70-80 % with constant 30fps (maybe a drop sometimes but nothing bad)

 

The only thing I noticed is that the runway lines and aircraft lines are more jagged, I think due the lower AA setting but I can live with it

 

What would be the settings for MultiSamplesPerPixel= MultiSampleQuality= If I set one bit lower?

 

Thanks for taking your time and helping me out.

 

I will post my further results tommorow.

Interesting. Thanks for your suggestion, Steve - I'm using the same settings as Eef, and they give me pretty good AA - acceptable to me, at any rate, around one of my favourite spots - Anacortes: for some reason, it is quite a hog, but now it's trucking along between 28 and 30 continuously - so it does work, and so I have added it into the fsx.cfg setup portion of the "How-To" doc.

 

Once again - Many Thanks! and thanks, too, Eef - for your patience and your input here. We need all the feedback we can get!

 

pj

So here I was all smug with getting DX10 settings all dialed in with decent AA (using 4 by 8 MSAA & 4xGSAA) and everything was beautiful using my TrackIR. But then I flew into cloud cover, using REX 2 clouds, and WHAM there goes almost 50% of my locked 30 fps and my GPU pegs 100%. So I reran the same scenario using DX9, with 8xS and 2xGSAA, and the FPS drop bounced between 3 to 5 fps with the GPU at around 70%; the clouds' visual quality was as good as in DX10. So I'm still on the fence about using DX10 preview, but that's OK with me and I'm not complaining (really). I'm just providing some feedback.

Did some research on REXE clouds and compression, and I reloaded a new cloud set making sure DXT5 and 1024x1024 were set (I find the 4096 clouds are too inconsistent looking). Using DX10 preview with the new cloud set loaded, the dramatic fps drop was gone, which is excellent! VIsually I can't notice any difference over what was loaded earlier. Also using DXT1 water compression is recommended to increase performance, though I've yet to do direct comparisons on this (but currently I am using this setting as well and the DX10 water looks great and fps is solid). Now back to flyng...

I understand that in dx9 when AA was set in app controlled in nividia inspector, fsx excluded the clouds from AA. Is this the same in dx10?

 

The reason I ask is that even with inspector set to app controlled and a fresh .cfg, when I even look at a single cloud in dx10 with in game AA turned on my fps plummets from 25 to around 10-12 which is not fun. I even have REXE dxt5 512 clouds installed. However, by turning AA off, clouds do not have any impact on fps. This seems to say that under dx10 clouds are anti-ailised no-matter what, which is unnessecary and obviously will kill a weak gpu like mine due to the complexity of the clouds.

 

So basically can clouds be excluded from AA in dx10?