Anti-Aliasing: irrelevant in the future?

Something is wrong with this thread; it belongs elsewhere (if only for Simon and Marco's contributions).

PS. I agree with Marco's logic. Especially :

 

I think we are moving away from the need to have FSAA because of increased polygon counts in models and procedurally created surfaces on the polygons. When I see FSAA most of the time it isn't really anti anything it's more of a blur filter with different degrees of accuracy. For me 1600x1200 on a 19 inch is enough to remove the need for it. To completely satisfy me 1600x1200 on a 19 inch with a 2x multi sampling is enough.

Texture filtering is I think more important as you increase resolutions. Not only that, but texture sizes effect visual quality the most in a game as you expand the resolution. A ten thousand polygon model with procedural lighting and culling maps, can be ruined by a low resolution texture, or even a high resolution texture that doesn't look realistic. Moving from models to sky maps and terrain textures or level textures is where it is even worse. Detailed textures of a simple pattern on a very high resolution texture that is stretched to fit the terrain which makes it seem low resolution is another problem. You have the base texture then another texture over that to add details which often results in seeing the tiling of the detail texture.

Megatexture is some new idea to improve this, but I don't know the details of how it works exactly.

 

You have got it reversed ... those things make MSAA less effective, but they actually increase aliasing. The only general way to filter anything procedurally generated is by supersampling (depending on the method you might be able to build it into the generation, but that's not going to be true very often).

 

Here's an idea. Would it be possible to manufacture a monitor (lets say a 24" monitor with standard resolution of 1920x1200) that would accept a higher resolution (in this case 3840x2400) then using a specialized chip resample that down to 1920x1200?

Assuming a video card can be made to run a game at acceptable framerates at this resolution would this...

1. Produce a less aliased image than running 3840x1200 [edit: I meant 3840x2400] on a 24" screen without resampling?

2. Produce a more "natural" and less "digital" looking image. IE - with a good enough chip would it effectly remove jaggies without unnaturally blurring everything?

3. Would it be just as efficient/almost as efficient to jus thave the video card do similar supersampling as opposed to running it at 3840x1200?

Regards,
SB

PS - I just don't see anything coming in the forseeable future that would obviate the need for some form of AA. I have to sit at least 4-6 feet away from my 24" monitor before Jaggies become un-noticeable at 1920x1200. However that does doesn't remove shimmering along edges of polygons that are in movement. For that I need to be 6-8 feet away fro the monitor at which point I only really see it if I'm really trying. However at that distance text also becomes unreadable in the vast majority of games, so that isn't an option either.

 

that would be a strange way to achieve 4x ordered grid supersampling, which sucks anyway (rotated grid is the way to go)

 

True, however, this would make it truly game/GPU independent. Meaning that assuming your video card can run at said "super sized" resolution you would have at "least" that amount of SSAA available to you regardless of what a 3D engine decides to allow you to use, or regardless of what a GPU designer decides is appropriate for their card.

Regards,
SB

 

He means that ordered grid supersampling really really sucks. I'd much rather have 1600x1200 unantialiased than 800x600 OGSS. With your example, I'd rather just buy a quarter speed video card (for way less $$$) and play at 1920x1200 without AA.

However, rotated grid is so much more effective and MSAA is so much cheaper that I don't play without it nowadays. Most of the time, enabling 4xAA and going down one resolution step actually gives you faster framerates.

 

Why is this so? Not saying that rotated isn't better (of course it is) but I would think that even if it's ordered, at least you get some kind of filtering that you otherwise wouldn't have. Am I missing something?

 

I realize that OGSS is not as good looking as RGSS, however that wasn't what I was asking.

My questions were,

1. Would 3840x2400 resampled down to 1920x1200 on a standard 24" monitor (using some custom chip in the monitor) show less aliasing and shimmering than a straight up 3840x2400 non-AA'd game in motion?

Considering that I expect faster video cards before I expect monitor makers to produce high pixel density LCD's, SED's or whatever, this might be a good compromise.

Also, I would prefer some form of SSAA even OGSS over no-AA.

As I stated earlier, while it has dog slow pixel response, I wish I could get access to IBM's 22" LCD with 3840x2400 resolution so I could test it out myself.

Regards,
SB

 

Why do it on the monitor? You're quadrupling your DVI bandwidth requirements for nothing.

 

our point is, the geforce 1/2 and radeon 8500 did this at the framebuffer stage (before leaving the card).
it was not stellar and very expensive. if you have a NV card that method is still available (under D3D games), using rivatuner or nhancer to set it : the 1x2, 2x1, 2x2 (what you propose), and 4x4 which is a bit insane (16x supersampling! that one looks great. but very inefficient off course)