- 488
- Germany
mipmaps are needed to reduce the jitter!For AC, yes, definitely. Like Masscot said, mipmaps are generated automatically by the engine, it's just a waste of space. Well, from my experience it is, adressing customized mipmaps doesn't do anything, so... I don't even talk about loading times in fact, I don't think it makes much a difference (supposedly longer, but by what margin), but it does make one in size, and size do matter (wink)
It depends very much on the game engine/virtual engine you're using your texture with. For some it's mandatory, for others a waste of ressource (mostly because they already create them).
And I don't know why I quoted your message, anyway, I 100% agree with it
Who you're apologizing for ? No need to, for others I can't speak but personally I'm just glad if I can help making it better in any way. For the rest, decision and such, you and @Langheck_917 are in charge, You do it your way.
And yes, you do have a good PC42 cars on Nords and mine gets hotter than a volcano, with the fps of the original Tetris.
if you always use the original sized textures then the renderer has a big amount of different pixels to get with a small amount of position change.
Imagine you have a big texture in a distance and the renderer picks the pixel according to your camera vector and where the texture is been placed. and the camera just moves a little bit, maybe a 1/10 of a pixel in output resolution, but in the texture it is a movement of 3 pixels, it will output you a different color then. To solve this you can use anisotrop filtering, but this will go several times and read a bunch of pixels around the pixel's position and need to do an avarage of it. This costs time and GPU usage.
So mipmaps are reduced versions of bigger textures and they already have some kind of averaging in it. So the jitter is reduces, because the renderer will read the same pixel with little camera movement.
Also, such texture read processes cost time too. With mipmaps a GPU could cache results and if a read command at the same position is done again, it could output it much faster. For every read process, the GPU already needs to interpolate the pixel neighbors, because you request the pixel with float values.
So a texture render on distant objects is way more efficient with a mipmap, than to have only the fullsize texture, which you can monitor in fps or GPU usage.
Last edited:
