P3D Dynamic Lighting impressions?

[gboz 157]

1 hour ago, him225 said:

Performance drop with DL depends on the quantity and complexity of the illuminated objects,

I don't know if my experience with DL and A2A planes proves or disproves that. :)
If I turn on A2A landing lights in mid-air (SSAA and completely clear weather at night) frame rates get cut in half, like 60 goes to 30. I guess there is a lot of air up there but is it really complex?

gb.

[Guest]

7 hours ago, AnkH said:

How can you disable DL specifically for an airport? Via AFD?

Most (not all) airports that support DL come with a configuration panel that allows you to enable/disable it's DL lights (sometimes several DL lights options), other vendors have a readme.txt file or help file that covers how to rename a BGL to enable/disable DL lights for their airport/scenery.

8 hours ago, him225 said:

don't get why LM decided to hold back shadows for them

As you can see, DL "as is" is already a hot performance topic that is very frequently brought up ... adding shadow processing for each and every DL light source (cap is 512 DL lights per scene) would be an impractical feature even for future hardware ... rendering would be in seconds per frame and not frames per second on the best of the best GPUs.

Cheers, Rob.

[Guest]

6 hours ago, gboz said:

If I turn on A2A landing lights in mid-air (SSAA and completely clear weather at night) frame rates get cut in half, like 60 goes to 30. I guess there is a lot of air up there but is it really complex?

The complexity doesn't change based on your virtual location (on ground or in air) ... the rendering of DL still has to go thru the same calculations per pixel to determine what (if anything needs to be adjusted for DL cone).  A positive adjustment to a pixel or an no adjustment to a pixel, still requires the same calculation.

Sounds like the A2A light cones are very large and/or have a long distance. Your view angle relative to the light cone will also increase or decrease the number of pixels that need to be calculated ... the more the light cone covers in a rendered frame the more calculations are need and the lower the FPS.

Cheers, Rob.

[him225 51]

18 hours ago, Nemo said:

Which entry should this be? I cannot find any.

MAX_POINT_LIGHTS=
MAX_SPOT_LIGHTS=

16 hours ago, gboz said:

I don't know if my experience with DL and A2A planes proves or disproves that. :)
If I turn on A2A landing lights in mid-air (SSAA and completely clear weather at night) frame rates get cut in half, like 60 goes to 30. I guess there is a lot of air up there but is it really complex?

gb.

It could you are comparing performance drop in air to when on ground which could be different since cpu bottleneck is likely to already limit fps on ground shadowing much of the drop from DLs.

10 hours ago, Rob Ainscough said:

As you can see, DL "as is" is already a hot performance topic that is very frequently brought up ... adding shadow processing for each and every DL light source (cap is 512 DL lights per scene) would be an impractical feature even for future hardware ... rendering would be in seconds per frame and not frames per second on the best of the best GPUs.

Cheers, Rob.

Yes it would, but a separate cap could be added for shadows as well and set at 5-10 or such.

 

 

[ahsmatt7 1,199]

10 hours ago, Rob Ainscough said:

The complexity doesn't change based on your virtual location (on ground or in air) ... the rendering of DL still has to go thru the same calculations per pixel to determine what (if anything needs to be adjusted for DL cone).  A positive adjustment to a pixel or an no adjustment to a pixel, still requires the same calculation.

Sounds like the A2A light cones are very large and/or have a long distance. Your view angle relative to the light cone will also increase or decrease the number of pixels that need to be calculated ... the more the light cone covers in a rendered frame the more calculations are need and the lower the FPS.

Cheers, Rob.

Is there any way this process could be optimized?

It doesn't seem like it's groundbreaking tech. I don't understand why it's so limited in p3d.

[Guest]

14 hours ago, ahsmatt7 said:

Is there any way this process could be optimized?

Yes, per my information above, you shift the order of rendering to have lights operate at a more optimal point that requires less complex light calculations (aka deferred rendering).  BUT, in doing that, you compromise other visual aspects such as transparency and AA (and a few other things).  P3D is forward rendered, XP11 is a deferred rendered ... you'll find all kinds of debates about which is better.  Both have their strengths and weaknesses.

For VR, forward rendering is currently better as it provides improved performance over deferred ... BUT as GPUs continue to improve in their capabilities/performance that might become less of an issue.

Here is a very good article about the pros and cons of deferred vs. forward rendering pipelines: https://docs.unrealengine.com/latest/INT/Engine/Performance/ForwardRenderer/index.html

Forward rendering in the situation outline in the article above provided about a 22% increase in FPS, but again it'll depend on the scene being rendered.

I'm not sure what you mean by "groundbreaking tech", "light" is the most complex aspect of any simulated 3D environment.  Light is both a wave and a particle and that takes us down the road of quantum mechanics ... to present such concepts in a 3D virtual matrix of X,Y,Z coordinates in a physical 2D space (monitor X,Y pixels) is definitely not trivial.  If you're interested in just how complex it is to implement light in computing, I recommend reading a really good book by Matt Pharr, Wenzel Jakob, Greg Humphreys on Physically Based Rendering (PBR) ... it will expose some of the concepts and mathematics involved, very interesting reading if you want to understand some of the science/tech involved.

Cheers, Rob.

 

[SAS443 1,716]

On 2017-12-11 at 4:24 PM, Rob Ainscough said:

adding shadow processing for each and every DL light source (cap is 512 DL lights per scene) would be an impractical feature even for future hardware ... rendering would be in seconds per frame and not frames per second on the best of the best GPUs.

Some engines have already solved this. With a caveat.
As long as spot and point lights are stationary, shadow map caching method is implemented. No performance penalty at all, unless the rendered object is moving.

Would work very well for city-scapes and airports (all the light poles etc). So the only headache left is that little airplane thingee..... 

[Guest]

Anything with a moving light source (not just own aircraft but vehicles, beacons, AI aircraft, AI road traffic, AI boats, etc.).  And then the issue of light on light on light on light and how that would impact shadows from each light source (even fixed ones) as an object travels within scope of the source.  Most airport lights will overlap and produce a different shadow based on number of lights and radius/intensity/distance of each light ... and that shadow for each light source will shift as the object moves thru it's cone/sphere of influence.

I'm curious how they solved this without compromises?  IMHO, when it comes to rendering lights in real-time, it's always a compromise no matter what techniques are used.  Only CGI movie quality rendering can afford the luxury of rendering a 10 second sequence for 30 FPS motion in hours or days. 

But with that said, one can still make it look pretty good even if not accurate to real world.  But in terms of optimizing DL in P3D, shifting to deferred from forward is not trivial and the results maybe good in some situations but bad in other situations (for example, haze in P3D looks far more realistic than the haze in XP11 which washes out all content color) ... always a game of give and take.  But like I suggested, as graphics performance improves and screen resolutions increase (less of a need for AA) the long term future (think 10 years from now) will slowly but surely shift to deferred rendering.

Another 2 cents.

Cheers, Rob.

 

[SAS443 1,716]

8 minutes ago, Rob Ainscough said:

I'm curious how they solved this without compromises?  IMHO

I'm no expert in the field, I just use the map-maker  for Unreal engine sometimes so I don't have in-depth knowledge. For Unreal Engine go here:

https://docs.unrealengine.com/latest/INT/Engine/Rendering/LightingAndShadows/LightMobility/DynamicLights/

Again this is a total hypothetical discussion, for what it's worth, it might not even be applicable in the World of flight sim, let alone in the  "old" ESP-engine...