Realtime Dynamic GI + Reflections + AO + Emissive plugin - AHR

[Voodoo]

Its happening...guys its happening!!! WOOOHOOOO!! ;D Ryan your a rendering wizzard!! Thx for the hard work its looking amazing!!

[TAN_]

Great work Ryan.

Let me help define " large ares" .

For example 16 km square or 32 square kilometer would be more than enough. Even whole skyrim world was something like 29 km square.

Keep up the good work mate.

[Errvald]

Great work Ryan.

Let me help define " large ares" .

For example 16 km square or 32 square kilometer would be more than enough. Even whole skyrim world was something like 29 km square.

Keep up the good work mate.

The human eye/camera can't see details after 3km anyway.

[RyanTorant]

The human eye/camera can't see details after 3km anyway.

Well, let me do some math. Lets asume we are using a movable Voxel grid, which is quite similar to what I'll have at least for the moment. If we are using a 3 square km area, we get a quad of sqrt(3) size, about 1.732 km side. If we are using a 512x512x512 grid, we get that each Voxel is 1732 / 512 m size, about 3.3 m. This is a bit unrealistic though, as normally you don't need a map that goes 1.5 km up and 1.5 down from the camera. I assumed a cube as bounds. With a proper bounding box, that could be reduced a lot, and I'm not even talking about cascades. Tracing performance is mostly the same, what limits outdoor scenes is the voxel resolution, and the voxelization time, as you'll need to get all the things inside that 3 square km area trough the pipe, at least once. For now, and actually that will stay like that for a while, everything in the bounds gets voxelized each frame

[TAN_]

The human eye/camera can't see details after 3km anyway.

Yeah, for long distance objects you use LODs, and if they are too far, like say 3km or more, you simply don't render them at all.
This is the trick CryEngine uses for trees and foliage for example

[KoldKam]

You are doing a great job bro. I can't wait to see it and test it out when it's finished. Keep up the good work!

[Daedalus51]

Awesome to see something again Ryan! And it looks gooood Keep us posted and of course...keep up the amazing work!

Cheers!

[Sitrec]

Yeah, for long distance objects you use LODs, and if they are too far, like say 3km or more, you simply don't render them at all.
This is the trick CryEngine uses for trees and foliage for example

That's pretty much standard, I don't think there is a larger, released game out there which doens't use LOD and distance culling.

[TAN_]

Yeah there is also imposters approach, bascially before culling instead of culling replaces the geometry with a polygon with alpha texture of a tree. Simply putting up "cardboards" that are very cheap.

Well there are lots of approaches in fact, some even include sophisticated algorithms of blend,ng foliage and terrain together or something. Also procedural generation is an option but depends on the game type and design actually.

[RPotter]

Yeah there is also imposters approach, bascially before culling instead of culling replaces the geometry with a polygon with alpha texture of a tree. Simply putting up "cardboards" that are very cheap.

Well there are lots of approaches in fact, some even include sophisticated algorithms of blend,ng foliage and terrain together or something. Also procedural generation is an option but depends on the game type and design actually.

Those all still fall under the category of "Level-of-detail"

[RyanTorant]

Just gonna drop this here... Remember that this is a WIP, and also, the voxelization stage is kinda like a brute force aproach for now. Everything gets voxelized every frame to a large, world covering voxel grid. Keep that in mind














... See you guys tomorrow

[JordanN]

It's better than lightmass quality but in real time. Insannne!!!!

Also, you should include screenshots taken in game mode. The editor uses up more resources so you should actually get a higher frame rate running without it.

[iniside]

On your hardware, with your resolution, and with editor, the results are I dare to say.. very, very good!

Now with optimizations, it might be even better!.

[Cube2222]

Are you planning to release a public alpha version? Can't wait to try it out!

[katoun9]

RyanTorant:
So what I understood from your posts is that your technique is like octree-based sparse voxelization cone tracing real time global illumination, but without the octree (so far ) and with ray tracing instead of the cone tracing (because it's cheaper)