guys, i really want to have cloud shadows, cockpit shadows and volumetric fog.

 

Yes

guys, i really want to have cloud shadows, cockpit shadows and volumetric fog.

i have gtx660ti, 2500k with 4.3 ghz. can my pc handle it with this features?

If you've been reading here and at the P3D forum very much, people are already talking about 2, 3 or quad SLI to cope w/ what 2.2 brings.  Perhaps you can add a 2nd 660Ti and get somewhere near what you will need.  As you must know also, you can dial back settings and lessen eye candy and still enjoy some of what 2.2 offers.  I have a SB-E hexacore running at 4.4Ghz, a Titan GTX overclocked, and when flying thru dense clouds in more complex scenery frame rate and lack of smoothness become show stoppers.   The cloud shadow implementation seems to have added a giant performance burden when in areas less than rural, and w/ heavy cloud cover, esp at sunrise/sunset which has always been my go to time to fly.   Of note, I am running the patched version of 2.2, however others bemoaning poor performance have stated installing the full version didn't solve these issues.  For me, 2.2 has really not been a positive experience so far.  I am also already tired of thinking about reinstalling add ons every time a new release comes.  Good time to be patient and see where things go.  I am very very loath to pick up a 2nd Titan until there is ultra solid evidence this will truly impact performance in a very big way.   Go to P3D's site and read some of the underwhelming reports from someone who has SLI'd Titan Blacks ($6000 for two of these).   Beau, an LM dev says 50% improvement in frame rate i assume will come w/ SLI.  I'm very certain things will improve in terms of the very nasty stuttering some are having in heavy clouds despite ok frames, so as I say patience is important as there are new issues IMO w/ what 2.2 has brought.  

 

I have a SB-E hexacore running at 4.4Ghz, a Titan GTX overclocked, and when flying thru dense clouds in more complex scenery frame rate and lack of smoothness become show stoppers.

 

Sounds like you're on a mission of deterrence Noel (this thread and others).  Welcome to world of shadow calculations ... and I think some are requesting the addition of "light rays" too ... so two Titan Z's (4-way SLi) "might" work here

 

 

 

Go to P3D's site and read some of the underwhelming reports from someone who has SLI'd Titan Blacks ($6000 for two of these).

.

 

Titan Black is $1000 - two is $2000 ... the Titan Z is $3000 each (which is essentially 2 Titan Blacks in one card 2-way SLi).

 

I think part of the expectations issues revolve around:

 

1.  Motion picture industry and their visual effects/animation

2.  3D shooters that have a very scripted an managed "small world" environments with very limited control over object count

 

Just in case some people aren't aware, visual effects/animation produced for movies is NOT performed real time and is rendered on server farms (think 100's to 1000's or more CPUs rendering complex light calculations/animations at frames per minute) with "baked" sequences.  Light and it's interaction with an environment is one of the MOST calculation intensive process (and it's what separates good 3D animation products from cheap/bad 3D animation products).  Visual effects software does not perform those calculations in terms of how visual light is presented and captured by the human eye, there are also no calculations of material absorption (the real physics behind why we have color), deflection, atmospheric influences, etc. etc. ... they are usually just object/texture properties with arbitrary values to represent how light will affect a material/texture -- in the 3D computing world one can actually bend the laws of physics to produce a visual experience not possible.

 

As for 3D shooters, take a very close look the next time you play a 3D shooter ... notice the background is a faded static blur?   You can't suddenly move up 5000 ft into the clouds, you can't suddenly run 1100 kts for 100 miles, your world is VERY limited to a managed polygon count.  Flight and water segments in these 3D shooters are again very control environments ... you move to the "next level" in the 3D shooter and you wait, often presented with some type of cinematic (pre-rendered) segue.  There is no gradual motion of the sun over time, there are no moving cloud shadows, there is no "I think I'll go to Athens now", the weather is always the same, etc. etc.  

 

Please review Beau's comments here: http://www.prepar3d.com/forum-5/?mingleforumaction=viewtopic&t=6363

 

 

Frame rates based on camera and sun angle are likely related to shadows. (especially if terrain/cloud shadows are enabled and longer draw distances are selected. The shadow map views render all the objects that can be seen in the main scene, or that might be between anything in the main scene and the sun. When the sun is on the horizon and you're looking towards it, it has to draw a lot more data into the shadow map views.

As for the for the original post which was has shadows disabled, there can be large amounts of point lights depending on the airfield and there will be more pixel overdraw down on the ground at that angle. How they perform may very based on certain things such like:
- GPU
- Resolution
- AA method being used
- Volume Fog Setting

Could you provide a bit more info about your hardware and settings?

 

Beau Hollis
Rendering System Lead - Prepar3D® Team 

 

and possible solutions with compromises:

 

 

 

We can can put some kind of sun angle check in the shaders or c++ and early out of cloud rendering into the shadow maps. The trick would be doing it without a glaring visual pop when the shadows turn on/off. I think at that angle the smaller shadow cascades like the one what surrounds the aircraft is shadowed by every cloud between it and the horizon and most of them are full screen size in that view because it doesn't have perspective. The shadow views are 2048 targets to a full screen cloud is a 4 mega-pixel raster and you could have hundreds of those in the first couple cascades.

 

Beau Hollis
Rendering System Lead - Prepar3D® Team 

 

 

I don't know what SLi/Crossfire will bring to P3DV2.x yet.  But I do know that more visuals (which people are requesting such as "light rays" improved water, etc.) WILL require more GPU processing power.

 

There are ways to "trick" the eye in order to improve performance, but those ways don't always work (rotating clouds issue is one such compromise).  Trees are basically X's (two planes with textures) -- trick works for the most part until you get very close to them ... houses don't have floors ... objects are surface models not solid models ... etc. etc.  These are all compromises made in order to retain any semblance of frame rate performance.

 

I love flying dawn/dusk also, and I still do ... but I understand the limitations of my GPU and as I've said, right now I'm very much GPU limited not CPU limited.  Can LM do some more tricks to help out on fps performance, maybe, but it could be another compromise is visual quality.

 

Cheers, Rob.

 

 

The cloud shadow implementation seems to have added a giant performance burden when in areas less than rural, and w/ heavy cloud cover, esp at sunrise/sunset which has always been my go to time to fly.

 

 

The performance drop at sunrise / sunset is a bug causing issues on the on the CPU cycle side - LM are aware and working on it (can't find that thread at the moment).

 

Cloud shadows @ medium settings aren't much of an issue on my measley 680...

Well, no doubt I've overlooked some things. There are so many variables I think I was again fooled prematurely.  I set up a flight w/ certain settings in a low complexity area, fly into the high complexity area and voila awful performance no surprise there.   Here are the things done for the last few flights that seemed to have improved things:

 

1.  Eliminated NI completely--was using Rob's prior set of 4x/4x.  Still looks quite decent--it seems they may have tweaked the internal AA maybe?  Seems better than it was in 2.0 for sure.

2.  Deleted the shader cache after #1, then re-ran P3D.

3.  Terrain Shadow at zero now.  Was fine in low complex area set above zero, not good in high density areas I see now.

4.  Increased autogen & vegetation to Dense from Normal (didn't seem to hurt perf in the moderately complex area (FTX NCA Reno to Sacramento so what the hey).

5.  Shadow quality slider at Medium.  Again, had that on Ultra initially in the low complexity area so was teased a bit for sure.

 

A few other changes which I know have been mentioned elsewhere by others, but I think I mistakenly calibrated in the low complexity scenario so now flying the T Duke in and out of KRNO w/ cloud shadows at 30K things like quite amazing at about 40min after sunrise (HDR at 5 in config--may have made a nice difference as the sunrise had much more bloom like effects than I've seen for ages), and voila seeing 36 fps and smooth at take off and then up in the mid 40's and up as I was flying towards Tahoe.    Very, very nice looking.  I think getting rid of NI was a very good thing and learning of the impact of a few new sliders is helping.

 

Thanks for that on the Titan Black whatever that is.  With what I saw today, if Titan SLI added 50% to that, I would likely choke it down.

 

 

1.  Eliminated NI completely--was using Rob's prior set of 4x/4x.

2.  Deleted the shader cache after #1, then re-ran P3D.

3.  Terrain Shadow at zero now.

4.  Increased autogen & vegetation to Dense from Normal (didn't seem to hurt perf in the moderately complex area (FTX NCA Reno to Sacramento so what the hey).

5. Shadow quality slider at Medium.

 

1. Yeah, I don't use NI any more.

2. Always a good idea when in doubt.

3. I have those turned off also, they cut my performance in 1/2.

4. I'm currently using Veg = Extreme and Buildings = Very Dense ... didn't notice any difference in performance.

5. I haven't done much testing with this slider ... have it a Ultra right now ... I haven't noticed much of difference going from Medium to Ultra either in visual quality or performance ... but after changing this setting it might be a good idea to delete shaders.  I'm still not clear how the shader cache is maintain during flight (I see them building at 1st time startup after a shader delete) ... I know a cache is there to speed up performance (as is most cache processing) but that implies as I fly more area's of the globe, my performance should improve over time as more will be cached.  Does deleting the cache mean this process will have to start all over again?  Is this why some people report very slow initial performance that later seems to "clear up"?  I'll be perfectly honest, I don't know how the shader cache is used and/or implemented during a session.  

 

I've looked at the D3D11SampleInlineRender.cpp provided by LM and can see where one can load custom shaders but I see nothing about setting cache or non-cache or how they are compared and resolved.  Maybe someone with more SDK experience can chime.

 

 

 

HRESULT D3D11SampleInlineRenderer::CompileShaderFromFile( LPCSTR szFileName, LPCSTR szEntryPoint, LPCSTR szShaderModel, ID3DBlob** ppBlobOut )
{
    HRESULT hr = S_OK;

    DWORD dwShaderFlags = D3DCOMPILE_ENABLE_STRICTNESS;
#if defined( DEBUG ) || defined( _DEBUG )
    // Set the D3DCOMPILE_DEBUG flag to embed debug information in the shaders.
    // Setting this flag improves the shader debugging experience, but still allows
    // the shaders to be optimized and to run exactly the way they will run in
    // the release configuration of this program.
    dwShaderFlags |= D3DCOMPILE_DEBUG;
#endif

    ID3DBlob* pErrorBlob;
    hr = D3DX11CompileFromFile( szFileName, NULL, NULL, szEntryPoint, szShaderModel, 
        dwShaderFlags, 0, NULL, ppBlobOut, &pErrorBlob, NULL );
    if( FAILED(hr) )
    {
        if( pErrorBlob != NULL )
            OutputDebugStringA( (char*)pErrorBlob->GetBufferPointer() );
        if( pErrorBlob ) pErrorBlob->Release();
        return hr;
    }
    if( pErrorBlob ) pErrorBlob->Release();

    return S_OK;
} 

 

and pulled this out of the documentation

 

 

 

Cloud Shadows and Shadow System ImprovementsThe biggest challenge we faced in shadow mapping the clouds is similar to the challenges we face with ground shadows. It stems from Prepar3D having very large viewing distances and the requirement to cover that view distance with enough shadow map texture coverage. It becomes a balancing act of performance and memory sacrifices to achieve acceptable visual fidelity. One big challenge of clouds in particular is clouds were not created with shadow mapping in mind and they rely on being blended into the image without writing depth. However, most modern shadow mapping techniques rely on the depth information to resolve if an object at a given depth is in shadow or not inside the pixel shader. These techniques require bounding volumes to be created around shadow receivers that project the range of these depth values between zero and one in DirectX applications. This could not be done easily for clouds, due to optimizations that are being done for the ground shadows bounding volumes, without losing quality on the ground shadows. So we’re required to use a multiple channel resource to determine things like the depth and alpha of the current cloud per pixel inside the bounding volumes. We’ve found ways of reducing ground shadow only memory and having cloud shadows on will be equivalent memory usage to before v2.2. Cloud shadows do come with increased GPU work, which comes from having to render the clouds into multiple shadow maps and evaluating that information per pixel for the scene. Because of this, cloud shadows may cause a FPS hit on lower end graphics cards, which may already be pixel shader bound, but should perform well on most modern graphics cards. 

 

Anyway, two Titan Z's and I'll be good to go!!  Just need SLi

 

Cheers, Rob.

 

 

 

Anyway, two Titan Z's and I'll be good to go!!  Just need SLi

 

Cheers, Rob.

 

Time to get a medical to re-activate my real-life PPL

 

al

Hi all,

 

Well, it's nice to know I'm not the only one who ran into a wall, and Rob again has done justice to tempering our 'expectations' to the 'realities' of REAL TIME RENDERING.

 

People thought I was insane to spread the work across 3 pc's and 3 GPUs, but I know from my X-Plane days this is a great way to uptick performance and smoothness.  

Sure, SLI would be a similar way to approach things, up till now, there hasn't been SLI in ANY of the 'big 3' sim platforms.

 

Rob is quite right about the 'server farms' - I've seen the ones at Pixar, and a relative used to be on staff there... it would take HOURS to render a frame... and they have many many pc's in an array to build their fabulous animations.  Of course this simply adds to our subliminal 'desires' and 'expectations' and we need people like Rob to bring us down to earth.  We all have the secret wish that our sims would rival the CGI motion-picture graphics we all enjoy in today's modern cinema!  I am guilty as charged.  Rob is right, that is NOT happening!

 

The more eye candy and complexity... well, it doesn't take a quantum physics professor to predict what will happen to frames --- they TANK.  We are like a crowd of spoiled children. Once we were happy playing PONG on a black-and-white screen.  Then Asteroids, Space-Invaders, Donkey Kong, and here we are sitting on our flight deck.  We all know what we WANT.  And we all keep trying to build it.  More add-ons, more hardware, real time ATC... sound effects, stick shakers, high-end seats...sigh...

 

Then the stuttering, and the break-apart on the multi-player monitor array... and ultimately heading back out the door for an alternative platform until P3D 'shakes out'.

 

REX has made a lot of noise talking about their upcoming Weather Direct 4 platform, and I have some high hopes for that.  But complex weather, hang-on high-detail airports, ground vehicles, runway textures, complex aircraft... the more you pile on... it's the law of inverse returns in frames... and eventually it becomes what one poster termed "a deal breaker", or put another way, unreasonable and unrealistic.  Then you are faced with trying to dumb things down to get back to how it ran before you piled all the garbage on it.  FSX on steroids.  P3D can do a LOT more than FSX.  But like all the sims - there are bottlenecks - be it CPU/GPU/VRAM or whatever.  And those bottlenecks?  NOT GREAT!

 

I can live without the ground shadows from the clouds.  Sure, they look nice, but they start to impact the performance, and I'm less enthusiastic.

 

Not knocking P3D in any way - I still feel it holds TREMENDOUS promise, and likely once everyone figures out what works and what doesn't - we can settle down and enjoy basic flight!

 

Thanks again for grounding our expectations Rob well-reasoned and perfectly postulated.

 

I am reminded of the old saying "how many whiskers make a beard?"  Another way of saying this, is at what point does the massive investiture in hardware outweigh the benefits of the simulation itself?

 

Occulus Rift holds some promise - but as Rob says, the shooter games have much narrower fields of vision and other restrictions that would not work well in airplane sims where you're looking 360 degrees out to the horizon.

 

Speaking for myself, $6000 for two video cards?  Not anytime soon folks.  That's just crazy - I'd rather pay that to fly RW PPL aircraft.  I had hopes that Microsoft's SERVER FARM could do for us what we couldn't do ourselves - give us BEAUCOUP rendering power without massive out-of-pocket expense.  That was the hope for MS Flight.  Sadly, it didn't work the way we imagined.

 

So we continue to 'wait'... and dream...

 

 

Anyway, two Titan Z's and I'll be good to go!!  Just need SLi

 

That's where I get off the bus!  Could do two Titans, but not two Z's.  I'm taking the CA Motorcycle Safety Institute course and am currently shopping light touring bikes so extra change of that magnitude goes for that ;o)

Well with SLI right now you want to go with Bog standard GTX 780's for SLI.

I have one already and have a deal on 2 used ones for $900 tomorrow. That will be 3 way

SLI that should tame P3D V2.2 Rendering requirements and move the bottle neck elsewhere.

I figure 2 GTX 780's will send the bottleneck back to the CPU but I am just guessing

Even my Mother board cannot do more than 2 - 16x PCI-e. For 3-16x PCI-e I have to upgrade

to a $450 Asus ROG motherboard.

 

Buying titans for SLI is IMO a waste of Money. They are over priced and won't performance gains

equivalent to the price. One Titan Z at $3000 will only move the bottleneck elsewhere: Dram, Chipset,

CPU. My 2 x used GTX 780's at $900 will deliver exactly the same performance at a fraction of the cost.

CPU. My 2 x used GTX 780's at $900 will deliver exactly the same performance at a fraction of the cost.

 

True but I already have one.  I guess I could try to sell it.  I wonder if my Titan GTX will SLI w/ Titan Black GTX?  May be able to find a used Titan GTX as well.  I think a comment from Beau was that you could expect ~50% perf bump w/ 2-way SLI.   Which means if you see frames of 16 in that scenario you're going to only get to 24, which is not bad but you do lose smoothness as you move down to the low 20's, even if it isn't real bad.   It's tempting to think you can stay ahead of the curve a little, but it doesn't last long for sure.   Been happy w/ the Titan just, as w/ SLI'd Titans, can't really run everything full bore--there's just too much scalability in the software.  Everyone has to compromise something.

I was sorta joking about buying two Titan Z's ... although, 3 years from now and I'm sure they'll be on sale for <$500 ... GPU industry still has plenty of growth left in it so don't hit the panic button just yet.  

 

I can almost guarantee you that 3-4 years from now people will be answering the "is my Titan Z enough for P3DV5.x" with most responses being "it'll work but you really need the" Europa series card to get the best out of P3DV5.x.

 

But my Titan Z reference was focused more on levels of end user graphics expectations that I've seen posted on various forums, just don't expect real time rendered Movie quality FX in a global flight simulation any time soon ... it's always going to be a hardware compromise ... the task is figuring out how to make the best of it and which blind eye to use

 

But like I said earlier, GPUs are just A LOT easier to scale moving forward than CPUs.

 

Cheers, Rob.