The old "Edge Anti Aliasing"

[TapamN]

What would be the reasons why you would not use all 6 polygons for a strip? Or is that done by the TA automatically?
I mean I understand models back then would not be as complex as today by far... but it still seems to me that such small strips would rather be an exeption.
So why did you take 3 as an average?

Not every triangle drawn can be part of a substrip of 6 triangles. You're going to have strips that can't be 6 triangles long, like having to draw quads for UI elements, which will bring down the average. If you have a strip of 7 triangles, it would have to be broken up into a 6 triangle substrip, and a 1 triangle substrip, effectively giving you a substrip length of 3.5 triangles.

I picked an average of 3 triangles per substrip as an arbitrary pessimistic value.

And what would be the disadvantage of using only 16 bits precision for UV coordinates? Misalignment/shimmering artifacts?

Misalignment. The 16-bit float format used by the PVR just drops the bottom 16-bits of an IEEE 32-bit float, so you only have an 8-bit mantissa (including the implicit 1 bit) . You only have enough precession to whole texel accuracy for 256x256 textures. If the texture is 512x512, the coordinates aren't precise enough throughout the whole texture, and could require rounding the texel position (e.g. (383/512) would have to be rounded to (382/512) or (384/512))

Also, it seems a lot of PS2-games have an awful lot of texture shimmering. I wonder why that would be, as the hardware supports mip-mapping.
The "standard" answer you would get would be they try to save space in the 4 MB eDRAM to make texture "caching" easier; but I guess that should not be aproblem if you schedule the GIF-bus transfers of textures accordingly...
I think I read here somewhere that the functions to select the correct texture LOD were very simplistic on the GS and would need manual touch-up in the game's engine.
Don't know if that's true, but there must be a reason why many PS2-games have this problem.

From what I've heard, PS2 mipmapping doesn't factor in perspective when selecting LOD, and always assumes the texture is perpendicular to the camera.

Compare Le Mans on Dreamcast and on PS2, for example.
PS2: Textures in the distance very sharp, but shimmering.
DC: Textures in the distance too blurry (no Anisotropic Filtering used and perhaps conservative texture LOD), but no shimmering.

First time I've ever seen someone complain about DC Le Mans's filtering. It's far better than the average DC game, and likely uses the PVR's aniso filter. You can open Le Mans's resources (modified zip file) and see text files describing per material settings that control either detail bias or aniso (it's just a number, so I'm not exactly sure what it means).

 

[jlippo]

From what I've heard, PS2 mipmapping doesn't factor in perspective when selecting LOD, and always assumes the texture is perpendicular to the camera.

It seems to be the case in at least games like rogue galaxy.
Not sure if one can adjust mipbias etc.

Aliasing it caused is quite horrible.

 

[Omarz]

Not every triangle drawn can be part of a substrip of 6 triangles. You're going to have strips that can't be 6 triangles long, like having to draw quads for UI elements, which will bring down the average. If you have a strip of 7 triangles, it would have to be broken up into a 6 triangle substrip, and a 1 triangle substrip, effectively giving you a substrip length of 3.5 triangles.
I picked an average of 3 triangles per substrip as an arbitrary pessimistic value.

Ok, makes sense.
As your calculation did not include any fancy stuff like specular lighting or modifier volumes, that again balances this a bit I guess.

From what I've heard, PS2 mipmapping doesn't factor in perspective when selecting LOD, and always assumes the texture is perpendicular to the camera.

Read that too... shouldn't that mean textures directly in front of the camera look better than the one on the sides?
Anyway this should not be hard to fix on a software-level, assuming you have a bit of control over it.

First time I've ever seen someone complain about DC Le Mans's filtering. It's far better than the average DC game, and likely uses the PVR's aniso filter.

It does? Would be one of the few games that do, I guess. But I think DC's AF is only 2x, so it would be of limited effect.
PS2's textures look sharper and a bit more detailed in the distance, pause at 5:27 for example:

This always bugged me a bit on Dreamcast. A few other examples:

https://imgur.com/csHM8qY

https://imgur.com/f3V7GIU

But don't get me wrong, I am allergic to pixel shimmering and aliasing, so I definitely prefer the Dreamcast look.

Only trying to say that it its not perfect on Dreamcast either and it has it's disadvantages.

 

[Cyan]

found a free program called Shader Glass. A CRT window for Windows that simulates a monitor. It is meant for emulators and classic games, but works on any active window, so I guess you can try it on any modern game?

It even works on youtube videos, so no more horrible looking Super Mario World videos on youtube.

ShaderGlass by mausimus

Retro shaders on top of any Windows app

mausimus.itch.io

 

[Omarz]

found a free program called Shader Glass. A CRT window for Windows that simulates a monitor. It is meant for emulators and classic games, but works on any active window, so I guess you can try it on any modern game?

It even works on youtube videos, so no more horrible looking Super Mario World videos on youtube.

ShaderGlass by mausimus

Retro shaders on top of any Windows app

mausimus.itch.io

Well for 2D-games like SMW we had things like xBR and stuff, which worked for some games... but I think we will see some great results from AI filters & shaders in the future.


For Anti-Aliasing, AFAIK the best method we can can use for recent games with halfway good quality/performance ratio to improve image quality would be DLAA (or DSR + DLSS)... still in the process of improving.

 

[Davros]

We also had ATi Smartshaders to emulate crt's for a small period of time

 

[3dcgi]

It you go past the 90s into the 2000s, Matrox Parhelia implemented edge AA. They called it Fragment Anti-Aliasing (FAA) and it only worked on triangle edges using an a-buffer like technique. It was like MSAA for edges and by being limited to edges it required less memory and had a lower performance impact than full screen MSAA.