66x Aa!!?

[Spud1]

http://www.humus.ca/index.php?page=3D&ID=59

looks interesting, i Havnt had chance to look at it properly yet but 66x effective AA seems interesting...

[Butcher]

Thing is you're massively into diminishing returns above about 8x. I'll bet most people couldn't tell the difference between 16x and 66x. Also it's way to slow to actually use in games.

[unreal]

Whoa look at the friggin smoothness on that!

[exAndrzej]

...66x effective AA...

When we launch CrossFire, we looked long and hard at the image quality available before labeling anything

Specifically, when we have 12x AA from the dual cards and then choose different centres to super sample, you could make the argument that you are offering 2 x 12 = 24x AA

When we actually inspected the results, the 'real' answer was 14x

i.e. stack pictures in a row with varying levels of AA and the one that 12x + 2x resembles is 14x

Not sure anyone has ever printed out every scene option from one to 66...

...but I suspect that the 'real' answer might happen slight before then

[exAndrzej]

Sorry - just thought of something else

With CrossFire, we are labeling on 'the quality of the actual output' rather than the straight maths because we think it is an accurate description of what you get

Wonder if 'anyone else' will get 'creative with the accounting' ?

[Trippledence]

Nice plug.

[iMc]

I cant tell the difference between 8xS on my 6800 with 1x in the program and 11x in the program.

[oralpain]

NVIDIAs 8xS AA is 4x MS and 2x SS. If you set 8xS in the drivers it forces those values and they normally cannot be overridden by any program, this is probably why you saw no difference. Try it with 4x AA, or manually set a high MS AA value with a program like rivatuner.

I'm going to see what 176x AA looks like (16x MS, 11x SS).

EDIT: Seems to work just fine, though I cannot see a different at the highest levels of MS+SS AA.

[suryad]

ARgh...just make grphx cards powerful enuf to run any games @ beyond 1920 res...no one will require AA then...

[exAndrzej]

...games beyond 1920...

I have been waiting for that for a while !

However, I don't think we wil get it just yet

There are several issues...

1) If I a remember correctly, the resolution you are looking for is really around 3000 x 2000 at which point there are so many dots that your eye should not see jagged edges at all

2) It would not be enough to simply have the screen at 3000 x 2000 because the images/textures would be all wrong... ideally you want the game designer to have created all of their assets at the highest possible resolution for close-up shots on a massive screen - and then 'scale down from there' for smaller screens/further objects

3) Costs will be prohibitive - both for the screens and the developers... you are talking about IBM T221 level screens for £399 (or less) and armies of artists and coders willing to work on minimum wage (ultra-talented folk create market leading games for 'burger-flipping wages') to create the game itself

Most likely scenario ?...

Some kind of major change in the technology required to create the screens we see in Arnold & Sharon's apartment (Total Recall)

PLUS

Move away from drawn art to the mathematical model of the world used in the top Hollywood pictures where people do not sketch textures - rather they are generated from mathematical procedures

BTW: If you are a budding graphic artist - keen on a career in game development - I have been assured that you will find a cool job at the end of your course... you will be standing over the shoulders of the mathematicians and programmers to make sure they get the numbers right !

[Kezzer]

But just think, it will happen one day

[Butcher]

NVIDIAs 8xS AA is 4x MS and 2x SS. If you set 8xS in the drivers it forces those values and they normally cannot be overridden by any program, this is probably why you saw no difference. Try it with 4x AA, or manually set a high MS AA value with a program like rivatuner.

I'm going to see what 176x AA looks like (16x MS, 11x SS).

EDIT: Seems to work just fine, though I cannot see a different at the highest levels of MS+SS AA.

They aren't using the AA hardware for their 11x so whatever driver setting you use will work fine and will combine with whatever hardware setting you have. Massive diminishing returns though.

Games beyond 1920 res requires screens beyond 1920 res - not easy to make them.

[oralpain]

The resolution of the screen itself is only part of the picture. A 3000x2000 reolution screen that is 2 meters across viewed from 3 meters away will still clearly have jaggies. A 300x200 resolution display that is 3cm by 2cm will look very smooth.

They aren't using the AA hardware for their 11x so whatever driver setting you use will work fine and will combine with whatever hardware setting you have.

You're right, I set 8xS AA and can see the diffences with the additional levels of SS AA the program applies.

Games beyond 1920 res requires screens beyond 1920 res - not easy to make them.

Most decent 21" CRTs can do 2048x1536. I play the original battlefield 1942 at 1920x1440.

At 2048x1536 AA makes little difference on my monitor as each pixel is smaller than the dot pitch of the screen so it's got a nice anolog blur to it, like a TV screen. Too bad refresh rate tops out at 65Hz at that resolution.

[exAndrzej]

...The resolution of the screen itself is only part of the picture. A 3000x2000 reolution screen that is 2 meters across viewed from 3 meters away will still clearly have jaggies...

Good point !

The back of your eyeball is probably around 4,000 x 3,000

There is a combination of size and distance for a 3,000 x 2,000 screen that should keep your retinas completely occupied

[Rys]

Thing is you're massively into diminishing returns above about 8x. I'll bet most people couldn't tell the difference between 16x and 66x. Also it's way to slow to actually use in games.

That's a fairly false statement, bar the performance part. You can still see edge aliasing at almost everything bar certain angles, with multi-sampling, at levels upto and including 16x (and beyond).

Think about sampling frequency and where the ROPs take their sub-pixel samples from (I'll explain it if enough people want to know) for the reason why.

[Rys]

There's also the issue of fragment shader-introduced aliasing to consider, multi-sampling schemes might not get rid of (the aliasing isn't in geometry, but the colour values output by the fragment hardware).

AA isn't an easy problem to solve!