[H]OCP does Radeon9700 I.Q....

Here's an illustrating example of why gamma correct AA is important.

http://hem.passagen.se/emiper/gammaAA.html

I've rendered a very simple white against black edge with 6x SSAA ala Radeon 8500. If you examine the image on the left you'll see that while the image certainly is better than a non-AA shot would be you can still see jaggies even though I used 6x AA. In the image on the right I changed the gamma to 2.0 use paint shop pro. Now look how much smoother it looks when we moved the linear averaging into a proper linear space. An even better way to see it is to look at the left image while slowly raising you gamma towards 2.0-2.5, (whereever your monitors sweet spot is), as you're approaching the optimal gamma value you'll see how the edge slowly smoothes out and jaggies goes away

I must say that being gamma correct is just as important or even more important that taking lots of samples. A gamma correct AA of 4x probably matches or beats a non-gamma correct AA of say 6x or even 8x. I think gamma correct AA is going to be standard more or less in the future. I'd be surprised (and disappointed) if NV30 didn't have it.

 

Humus, am I to assume that you rendered those images in software?

I'd still really like to see a wireframe shot of the Radeon 9700...given what we've seen so far in this thread, the difference should be immediately obvious.

 

No, they were rendered in hardware on my Radeon 8500. What I've done is to move the linear averaging into a linear space, which is what gamma correct AA tries to do. I've added some line shots too, the effect is even more obvious on those.
For this reason I'm also very interested in seeing AA wireframe shots from a Radeon 9700, preferably with comparison shots from a GF4.

 

OT: About them BMP's...

PNG compresses better than a ZIP'ed BMP, there's no un/zip hassle, and it's still lossless. Brent, get yourself IRFanView* and start compressin'!

*Not sure if it's free for business use. It's a sweet program, though.

 

Interesting. May I ask exactly how you managed to do that on the 8500? Pseudocode would be nice...

Heh...hopefully somebody will listen to you Either nobody with a 9700 has read the later posts in this thread yet, or has just ignored me...I wonder if Donald would be willing to do a couple more shots?

Anyway, here's what I think the ideal shot would be:

1. Wireframe shot at the same position on the different video cards (should be easy enough with a saved game...).

2. Optimal situation would be one in which there are lines at may different angles, which is why I like Morrowind...but that won't work for a GeForce4, as the shots don't show FSAA (Which brings up an important point...it may not be so easy to capture FSAA on the GeForce4...).

3. Good comparisons would be between the 2x FSAA modes (which, presumably, use the exact same sampling pattern, and so the main difference should be in the gamma correction...), and 4x FSAA on the 9700 vs. 4xS FSAA on the GeForce4 (not the exact same sampling pattern, but more similar than 4x FSAA, which should help to isolate the gamma improvements).

 

I'd be happy to take the screenshots, but I don't have a 9700. Or a GF4. Oh, and I don't have Morrowind either.

 

Gee thanks!
We'd like you to try to draw what you imagine they'd look like though, anyways, ok?

 

They're probably too busy playing, I know that's what I'd be doing. 8)

Better get them to take those pics fast, UT2003 demo is coming out this week!

 

Oh, man, you'd better believe I'm getting that the day it's released!

And, fortunately, DSL should be up and running tomorrow! Excellent timing, if I do say so myself 8)

 

Exactly. 8)

 

That's what I am doing, ha ha ha !

 

Give me a wireframe-only command for a game I have and I'll do it (when time permits):

Q3A?
UT?
DungeonSiege?
Commanche 4?


Mize

 

Does anyone know if the 9700's AA being gamma correct will make as much of a difference with an LCD screen?.

I'm wondering because of reading articles about cleartype and LCD screens but that's probably just related to improving clarity as opposed to reducing aliasing.

 

Good God man!!! You can't go about posting stuff like this without some linkage for heaven's sake. I think I soiled my armor!

 

Thanks

I did some looking, and here's how you'd do it in Q3A:

\devmap q3dm1 (You could use another map...but q3dm1 just has one entry point...)
\r_showtris 1

 

ummmm for me any game in OpenGL that I try to take a screen shot with AA on looks like this on my R9700:

http://www.pc-gamers.net/jb/reviews/R9700/temp/shot00010.jpg

So you might want to try a D3D as those work find with D3D (of course maybe thats been fixed by now).

 

Ehm, I suppose you misunderstood what I did. There's nothing fancy going on here. I just draw a white quad, no more advanced than that. I also enable 6x FSAA. This in on a Radeon 8500, so no gamma correct AA, it just averages the values linearly. So the shot at the left is just that, a white quad drawn with 6xFSAA enabled. The edge doesn't look quite smooth though.

I then took this image into paint shop pro, and raised to gamma to 2.0, that's the image on the right. Now all of a sudden it looks very smooth. Why? Because with a gamma of 2.0 - 2.5 (depending on the monitor) you get linear output, which fit's well with the linear blending of samples done in the AA implementation.

We could of course always use a gamma of 2.2 or so in games and all FSAA implementations would look much smoother. Unfortunately, this will of course make everything look washed out though, because game contents are made with gamma 1.0 in mind. Now say a game were to be implemented with correct gamma from the start, and make textures such that they will look good at gamma 2.2. The problem that would now occure is that we'll get banding in dark areas of the image, and to get rid of that we'll have to use high precision textures, 32bit textures wouldn't do.

If we can use 16bit/channel texture and gamma of 2.2, then the problem would be gone. Not only that, we would also get better results of alpha blending and similar operations. In the far future this will probably be the way to go. This + independent gamma for each window and we'd fine.

In the shorted term though, ungamma the textures, then blend linearly and then gamma the result is probably a better method. This is because the eye's response to light is quite non-linear, similar to the output curve of a normal monitor at gamma 1.0. So for most uses still a gamma of 1.0 is the most optimal way to represent texture data.
This is one of the interesting features that will appear in DX9, the possibility to ungamma textures when sampling them.

 

Oh, that's all? Heh, I thought you actually did something within rendering.

Yeah, for just white on black, what you did should be the same as gamma-correct rendering...

Btw, one other thing. I'm not sure high-quality source art is needed for good gamma...just high-quality internal rendering (Assuming, of course, that the textures aren't too smooth...that will cause banding no matter how you render...).

 

It looks from jb's post like I'll need to use D3D. Anyone know how to go to wireframes only in UT, Shogo, SS, SS:SE demo, Giants or Commanche 4, GroundControl or Dungeon Siege?

I can put some screenshots up tonight if anyone can get me to a wireframe only output.

Mize

 

Heh, weird. What driver version are you on? This problem was fixed some time ago (unless it's a different bug all together), but I don't track which bug fixes go into what releases.