I have experiemented with this for a while and there are both advantages and disadvantages. Like your thoughts on this:

Pros:

Dramatically better LOD as there no MIP maps. Consistant appearance at all distances
Can be used strategically in the same NIF (one node use TGA, another use DDS)
No compression artifacts
No specular issues from normal maps

Cons:

TGA files are natively larger and must remain unconpressed
Requires models have vertex colors applied
TGA files will not display in CS (and produces ignorable wanings)
No normal mapping possible
No transparency or parallax with 24-bit TGA
Can't be used on rigged meshes

Most if not all of the pros can be done with dds as well or not?

1. You can save dds files without mipmaps...although I think that causes bugs. Not sure why tga files should cause less bugs though. Maybe you can make larger mipmaps as well with custom settings, never tried.

2. -

3. dds can be saved without compression (and thus without artifacts) as well. Not sure how an 8.8.8 RGB dds compares to tga regarding filesize.

4. dds doesn't need a normal map either unless it's a skinned mesh. In which case it probably wouldn't work well with tga either. In most cases normal maps are a must anyway since textures without normal maps aren't affected by game fog and dynamic lighting.

Never seen a DDS with less that (4) MIPs but I haven't tried that. Something to look at.



Something else to compare.



This can be added as a Con for TGA but I wasn't think of using TGAs on rigged stuff. Valid point though.



Did not test this and thats something I need to do. I do know that TGAs seem to react better to HDR than DDS, probably because blowout is coming off the alpha of the normal map though.

I think there's a contradiction here -- the purpose of mipmaps is to make textures look better at a distance. Without them, you get http://www.shinvision.com/wp-content/uploads/2009/06/mipmapping-egz.png on distant surfaces because the GPU is taking the color of the single texel at the center of the pixel, rather than the (approximate) average of all the texels within the pixel's field of view.

Graphics cards can generate mipmaps automatically when loading textures, and I would guess that's what happens when you use a TGA file as a texture. The mipmap can't be manually fine-tuned by the artist in this case, though.

Didn't intend to try this on something like a road that has a tiled textured over foreshortened perspective. Issue is moot though.

888 is larger and the difference can be sometimes considerable as it all depends on texture content because TGA has no regular size like DDS has.

Texture aliasing without mipmaps vs. blurry results with using mipmaps.

I hate aliasing more.

Also from a technical point of view, mipmaped textures help performance a lot. And compared to aliasing, they are better quality too.

The blurriness happens when a surface is viewed at a shallow angle, such as distant ground textures in a first-person game like Oblivion, because the mipmap averages colors over square areas of the texture and that's not a good match for the area covered by a pixel at a shallow viewing angle (it's a tall narrow traqezoid). http://en.wikipedia.org/wiki/Anisotropic_filtering fixes it, by doing the averaging over non-square areas that more closely fit the pixel's field of view.

Unfortunately, Oblivion doesn't support AF natively, but you can enable it manually with nVidia's and probably ATi's driver control panels. OBGE might support it too; I haven't checked.