Re: Texture caching... amazing improvement in memory with Cycles, could be great in LuxCore!
Posted: Thu May 06, 2021 2:30 pm
Oh no, what I mean is way simpler, or at least it's way simpler in my mind, something that it's not always true in reality LOL
What I mean is that based on distance to the camera and size of the final output resolution maybe depending on the perceived texel size.
The resolution of the textures that will be used could be changed, since it's not real time it has to be done once per render, in that situation you could calculate some "simple" mip-maps that reduce the texture size based on distance, that way if an 8k texture appears just in one pixel of the screen it could be reduced.
It sounds like a mip-map description... but it's way simpler I think since you do a pre-pass doing a calc for all the required texture sizes and then you "replace" those textures with the reduced ones, all that could be done on CPU so memory is less of a problem and the textures you upload to the GPU memory are the reduced ones, internally it would be that they are just the textures, in the end both GPU and CPU would be using those reduced textures.
It's something like the "simplify" option used in Cycles for the texture size but it works with depth.
Ok, this may be a totally silly thing, or it may be exactly what you discarded and I may have not understood your explanation, but just wanted to clarify the idea I was proposing
What I mean is that based on distance to the camera and size of the final output resolution maybe depending on the perceived texel size.
The resolution of the textures that will be used could be changed, since it's not real time it has to be done once per render, in that situation you could calculate some "simple" mip-maps that reduce the texture size based on distance, that way if an 8k texture appears just in one pixel of the screen it could be reduced.
It sounds like a mip-map description... but it's way simpler I think since you do a pre-pass doing a calc for all the required texture sizes and then you "replace" those textures with the reduced ones, all that could be done on CPU so memory is less of a problem and the textures you upload to the GPU memory are the reduced ones, internally it would be that they are just the textures, in the end both GPU and CPU would be using those reduced textures.
It's something like the "simplify" option used in Cycles for the texture size but it works with depth.
Ok, this may be a totally silly thing, or it may be exactly what you discarded and I may have not understood your explanation, but just wanted to clarify the idea I was proposing