Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Jul 23, 2020.
60s on high preset in Novigrad, come on
It really depends on how the scaling really happens. Once Unreal 5 is out, we can actually see how they do it. Other engines might have different approaches. They mentioned multiple times that their max detail level is 1triangle = 1pixel for mesh LODs, I don't know enough on how they will do it for textures.
All the people involved in this and whose job this is, say that it will work this way. I have zero reason to dispute it, honestly.
You also realize that the "manual" method of pulling that texture, involves an algorithm that does that dynamically in the end. If you replace the extra texture with the ability to scale a single texture because you can fit it in memory or load it instantly, then it's basically the same thing, without the extra texture.
False. Texture is color representation of some object space on polygon. If you want to have accurate representation of 12 pixels on screen from 4096x4096 pixel texture covering such surface, you'll have to either load it entirely (from VRAM to GPU) and do proper downsampling. Or you will load few small chunks and throw away accuracy.
In this situation, loading and processing 64x64 texture makes downsampling an easy job and no data would be ignored as this 64x64 texture was generated by reduction that used all pixels on 4096x4096 texture.
That's common problem with texturing and issue of 16tap AF. If source texture is too low resolution, surface of object on screen becomes blurry.
If texture has too high resolution, then 16 samples per pixel is too few to accurately represent original texture after downsampling and that results in shimmering, moire patterns, or just pixels that change a lot as you look at them from different angle or distance.
Don't you remember what forcing -1 texture LoD globally did in past? Then -2, -3, ... ?
Except that graphics workload do not give a crap about resolution and screen refresh rate today, and this is since GPUs render anything different from a wallpaper. Same as texture resolution and mip-mapping, anything about to compute the required VRAM is just senseless, you may have tons of resources loaded in a default heap on the VRAM and you can get closer any mesh than a 1:1 viewport, and of course frame buffer are just a small portion of the used VRAM. So yes, talking about resolution, refresh rate and vram amount is just senseless, the only rational measurement is the obtained performance.
So... What is exactly false? Did you even understand what I wrote?
Right, so how are they doing it and why do they insist that there is only a single copy of a high resolution asset instead of multiple copies of the same assets at different quality levels?
Google yourself what mipmaping is. Because it is exactly thing you believe will no longer exist in PS5.
sampler feedback and mipmapping are cooperative peas in a pod, expect MORE mipmaps, not their removal.