GUIDE xLODGen Terrain Settings Compare

[TechAngel85]

@DY
Is z-fighting reduced or eliminated with TAA?

[DoubleYou]

Good question. If it is, however, TAA is rather blurry, and I think it may be incompatible with ENB.

[TechAngel85]

Good question. If it is, however, TAA is rather blurry, and I think it may be incompatible with ENB.

Yeah...I've heard blurry this, blurry that. I've been running it since day one and I have no issues with it. I think people just what to see the pixels on the distant objects and aren't actually playing the game much, but standing around a staring that things. I don't use ENB now, but I'm pretty sure I was running TAA with ENB a while back. Might do a test with it on/off.

[TechAngel85]

Removed "confusing" information from my post. See descriptive information above compares for what they are and included.

[z929669]

Removed "confusing" information from my post. See descriptive information above compares for what they are and included.

Yeah, that was my fault. I wasn't clear about some of the details behind the settings and mods behind the screens. It's all clear to me now

 

The gen time for your xLODGen Output in each of the relevant runs would be most helpful.

[DoubleYou]

@DY

Is z-fighting reduced or eliminated with TAA?

I checked this today. It doesn't eliminate it.

[DarkladyLexy]

hey, guys, I have look over your recommended optimal settings you use on STEP and I am curious to know why you guys don't create MipMaps?

 

I also note yourecommend BC7 Max (I must admit I don't know the difference between BC7 Quick and BC7 Max)

[TechAngel85]

hey, guys, I have look over your recommended optimal settings you use on STEP and I am curious to know why you guys don't create MipMaps?

 

I also note yourecommend BC7 Max (I must admit I don't know the difference between BC7 Quick and BC7 Max)

I asked the same thing and Z mentioned in the Discord that Sheson said the game doesn't even use the mipmaps, so I'm assuming there is no reason to generate them.

 

BC7 Max vs Quick is akin to saving as Fast or Fine. It's just the quality of the compression. There's likely not going to be any noticeable difference for LOD textures.

[DarkladyLexy]

I asked the same thing and Z mentioned in the Discord that Sheson said the game doesn't even use the mipmaps, so I'm assuming there is no reason to generate them.

 

BC7 Max vs Quick is akin to saving as Fast or Fine. It's just the quality of the compression. There's likely not going to be any noticeable difference for LOD textures.

i see that is interesting thanks Tech

[Greg]

I asked the same thing and Z mentioned in the Discord that Sheson said the game doesn't even use the mipmaps, so I'm assuming there is no reason to generate them.

 

BC7 Max vs Quick is akin to saving as Fast or Fine. It's just the quality of the compression. There's likely not going to be any noticeable difference for LOD textures.

I take it BC7 Quick vs Max is something along the lines of achieving a reasonable compression ratio really quickly or achieving a higher compression ratio that requires (perhaps significantly) more time?

[z929669]

I take it BC7 Quick vs Max is something along the lines of achieving a reasonable compression ratio really quickly or achieving a higher compression ratio that requires (perhaps significantly) more time?

I think this is true. Quality of quick may be theoretically better, but you will never see the diff if that's the case.

 

@Lexy

sheson has verified indirectly that terrain probably doesn't use mips for diffuse, and I tested that it probably does not for normals either (VRAM is pretty much unchanged with/without). Makes sense too, because terrain probably just doesn't need mips, and I don't see how there would be any benefit. The detail is really not as important as blending for terrain LOD.

 

Also, the settings on the wiki right now are probably overkill. The settings Tech found in a previous post seem to be just as good with half the size and time and slightly better performance. I'm still running a big set of compares under varying conditions, but the final result looks to be closer to Tech's settings. I even have some evidence that the xLODGen result may not be as good as just skipping this part of LOD gen ... at least with Cathedral Landscapes. Terrain LOD Gen seems to bork the otherwise nice blending of CL landscapes with landscape objects ... final determination coming soon in a post to this thread.,

[DarkladyLexy]

thanks, Z I will watch with great interest.

[TechAngel85]

I even have some evidence that the xLODGen result may not be as good as just skipping this part of LOD gen ... at least with Cathedral Landscapes. Terrain LOD Gen seems to bork the otherwise nice blending of CL landscapes with landscape objects

Keep in mind that xLODGen is for more than just blending terrain. It also updates LOD to match changes to the landscape from mods. For example, RWT changes the shape of the landscape across from Solitude along the marsh/river edge. Without xLODGen, you'd have mismatched LOD for those terrain edits. You can actually see this in my shot overlooking the marsh (image 1 vs other images in set).

[z929669]

Keep in mind that xLODGen is for more than just blending terrain. It also updates LOD to match changes to the landscape from mods. For example, RWT changes the shape of the landscape across from Solitude along the marsh/river edge. Without xLODGen, you'd have mismatched LOD for those terrain edits. You can actually see this in my shot overlooking the marsh (image 1 vs other images in set).

I get it ... just sayin' though.

[z929669]

Compare set 1: xLODGen Quality Settings Compare
 
First post of my fairly exhaustive testing. This first set looks at xLODGen quality setting differences between the latest 'optimal' settings and Tech's "Custom Settings". The main difference is that the 'optimal' settings use higher resolutions and quality = 0 (highest quality). Tech's revised settings use lower resolutions and slightly lower quality. NOTE: The revised settings used here are identical to Tech's settings with the exception of LOD32, where I set Quality = 4 rather than 0 as Tech has it. These textures are almost never seen except for when looking at the map (proof coming in a later post).
 
SPM stats are included in the images (you can do this in Windows using ALT + WinKey + Prtscn while SPM is running).
 
NOTE: all of these and forthcoming compares look at terrain LOD ONLY. Do not be fooled by crappy tree and object LOD - DynDOLOD modifications are NOT included here so that we can examine STEP under two different xLODGen quality scenarios. Also, do not be fooled by smoke and clouds. These are dynamic in some shots, so many diffs noticed when toggling between any given compare set are due to this texture interference in SOME compare sets.
 
These and forthcoming compares examine 9 different locations. For this set, we are only looking at two variants per location in the following order (including xLODGen generation time and output size):

1. 'optimal' settings                 36:22, 4.19 GB
2. >> Tech's revised settings    23:45, 2.91 GB

[spoiler=Compares]

Conclusion

Hopefully, everyone will agree that there is NO quality difference. There is also NO significant performance difference. Variants of Tech's settings will be used for the remainder of these compares (see following posts).