Reading this article got me thinking about the status of VWDs in Morroblivion.
Q1) How extensive are VWDs in vanilla meshes.bsa and what about custom ones for Morroblivion, like RAEVWD for Oblivion?
Also, I read from a post Zilav made that optimized VWDs were made according to the techniques mentioned in the above article for Skyblivion.
Q2) Are those only for vanilla meshes or also for all the RAEVWD meshes?
According to the article, combining meshes and textures for LOD meshes reduces drawcalls. I don't see why it shouldn't be done for standard (non-lod) meshes to reduce drawcalls for them too. But, then some meshes might require 8096 textures.
Q3) What is the max resolution texture the Oblivion engine supports?
Q4) If standard meshes and LOD meshes share the same texture, is there a need to reduce the LOD texture size, since the LOD would use a mipmap of the standard texture... or would they?

Thanks for sharing that
Thanks for sharing that article, junkacc. I think it does a great job of clearly explaining the performance issues of distant lands and statics and discussing the methods of optimizing performance.
Q1) The "Oblivion - Meshes.bsa" contains about 10,000 files in the \Data\Meshes directory. I did a search and found only 130 items that end in "_far.nif". These _far.nifs were all architecture for the major cities: IC, Bravil, Bruma, Cheydinhal, Kvatch, Skingrad, Daedric and generic castle, etc. Noticably, there are no rock or terrain meshes. Which explains why RAEVWD makes such a difference. Also of note is that the RAEVWD core subpackage only contains 142 _far.nifs and most of those are only for the main cities. RAEVWD starts getting visibly better once you add in the subpackages for Wilderness Architecture, Aylied Ruins, Imperial Forts and Large Rocks. The total RAEVWD mod contains 681 _far.nifs.
Q2) I don't know what the skyblivion method for VWD generation is, but here is a quote from the RAEVWD readme:
Regarding your comment on combining textures: many NIF files already share a single texture file or use a common texture resource. For example, many armor models will share portions of the same shiny metalic plate texture or chainmail texture. The biggest drawback in trying to combine more textures after the fact is that it will break the UV coordinates in the models. Of course, recalculating UV maps programmatically is an option. Coming from the Morrowind modding world, I still see the gold standard for automated Distant Land and Distant Object generation to be MGE-XE. You can set a size threshold to be used in selecting which statics will have LOD meshes generated, and you can set the amount of detail as a percent of the original to set the quality for the generated meshes and textures. Blender has some crappy decimation tools: there is one function that tries to remove vertices while preserving the original UV map (the results are models that may only be 10% smaller than the original), and there is another that can do a better decimation but breaks the UV map and does not recalculate it for you. Compared to the ease of use of MGE-XE, Blender is worthless for automated LOD mesh generation.
Q3) not sure, but the real question is can the average user's graphics card handle larger resolutions? Both in terms of maximum individual texture size and total VRAM for all textures. Keep in mind that memory requirements grow exponentially with linear increases in resolution.
Q4) The LOD mesh will take advantage of the appropriate sized mipmap, but the issue is that the full resolution texture with all mipmaps is what gets transferred across the PCI-E/AGP pipeline and stored in VRAM. The GPU hardware then calculates what pixel to paint based on your texture filtering algorithm (ex, trilinear will interpolate between two mipmap levels to create better quality textured polys). Bandwidth and storage are probably the major performance bottlenecks for high resolution textures (bandwidth for transferring the textures and VRAM storage of the textures). This, along with disk reading needed to load 100-1000s of megabytes of high resolution textures into memory also contributes to in-game stutter. Is this better than 100-1000s of textures of that are only 1-10 megabytes in size? Perhaps -- I think it would require testing and performance tuning to find the right balance.