I only mentioned resolution in reference to ideal scenario (ie. 1:1 texel coverage, which typically never happens in real-world outside of contrived scenarios).joker454 said:Someone else also mentioned resolution in reference to how much memory ideally is needed,
So no, I don't think it's dramatically important, because it's not even a real visual target anyway, just useful as a measuring stick.
Given this discussion isn't likely to be settled on any data proof, I'll just opt to agree to disagree here. I haven't done nearly the amount of work to claim expertise like some commercial projects have in this field, but I have done enough to know the concept can work - even in its simplest forms (eg. paging maps simply by their required mip-levels).Software solutions help, but I think we're still at the point where we just need more memory available to avoid visual fidelity loss.
8 years back I remember I've had similar discussions with people arguing over paging of textures from PS2 main->VRam, which really works on very similar principles if you want it done well. It seemed so obvious (I've seen savings of up to 10x on the bandwith) and people still argued against it.
But yea - if you want to truly exploit temp coherency - you have to make concessions on the data-structures, which does seem counter intuitive I suppose (since afaik most streaming systems try to pack data in ways that are closest to what HW will consume, and this actually requires the oppostite).
I am not debating the usefulness of more memory btw - just that there's still a lot we can do with what is actually available.