AA: which one looks better?
- Thread starter Ante P
- Start date
The left one is 6x, the right one is 4x ( good ole Windows zoom feature 
)
So, it's either a R300/R200 or it's a NV2x with the register hack.
As for which one looks best... I know I'm gonna look insane saying this, but... I prefer the right one ( 4x )
Why? Well, after zooming in real close, you realize that sometimes two nearby pixels are identical. That gives a very bad effect. On the 4x screenshot, that never happens.
Why doesn't it ever happen on the 4x screenshot? Well, the 4x sampling pattern is probably better ( If it's a R300, ATI stole the V5 4x pattern for it. But the V5 didn't have a 6x pattern, so that one is Canadian-made; it would explain a lot ) , and the differences between pixels are always smaller with 6x, so it's quite normal such things might happen more frequently.
I'd suppose that in most cases, 6x would look better than 4x, but in this one, it doesn't. Or maybe 6x is often worse than 4x, and we better wait for 8x or 9x to have acceptable performance?
Uttar
)So, it's either a R300/R200 or it's a NV2x with the register hack.
As for which one looks best... I know I'm gonna look insane saying this, but... I prefer the right one ( 4x )
Why? Well, after zooming in real close, you realize that sometimes two nearby pixels are identical. That gives a very bad effect. On the 4x screenshot, that never happens.
Why doesn't it ever happen on the 4x screenshot? Well, the 4x sampling pattern is probably better ( If it's a R300, ATI stole the V5 4x pattern for it. But the V5 didn't have a 6x pattern, so that one is Canadian-made; it would explain a lot
) , and the differences between pixels are always smaller with 6x, so it's quite normal such things might happen more frequently.I'd suppose that in most cases, 6x would look better than 4x, but in this one, it doesn't. Or maybe 6x is often worse than 4x, and we better wait for 8x or 9x to have acceptable performance?
Uttar
Babel-17
Veteran
In games a lot has to do with how the map is made. Some mappers go out of their way to minimize the nastiness of edge aliasing with the way they choose textures, colors and lighting. With those kind of maps you want to avoid any added blurring from your aa.
Other maps seemed to be designed with the premise that the user will have some kind of aa available or perhaps the mapper isn't concerned at all with what stark lighting and color/texture contrasts will do to the edges. With maps like those nVidia users can reasonably claim quincunx is a useful option.
Options like aniso and maybe adjustable lod bias increase your options to the degree you can use them. I just switched from a GF4 to a R9700 Pro so I'm still ooohing and aaahing over the differences. UT2k3 offers a nice cross section of maps for comparing the need and effectiveness of aa. No surprise of course that indoor maps generally need it less. Imo.
Other maps seemed to be designed with the premise that the user will have some kind of aa available or perhaps the mapper isn't concerned at all with what stark lighting and color/texture contrasts will do to the edges. With maps like those nVidia users can reasonably claim quincunx is a useful option.
Options like aniso and maybe adjustable lod bias increase your options to the degree you can use them. I just switched from a GF4 to a R9700 Pro so I'm still ooohing and aaahing over the differences. UT2k3 offers a nice cross section of maps for comparing the need and effectiveness of aa. No surprise of course that indoor maps generally need it less. Imo.
Uttar said:The left one is 6x, the right one is 4x ( good ole Windows zoom feature
So, it's either a R300/R200 or it's a NV2x with the register hack.
As for which one looks best... I know I'm gonna look insane saying this, but... I prefer the right one ( 4x )
Why? Well, after zooming in real close, you realize that sometimes two nearby pixels are identical. That gives a very bad effect. On the 4x screenshot, that never happens.
Why doesn't it ever happen on the 4x screenshot? Well, the 4x sampling pattern is probably better ( If it's a R300, ATI stole the V5 4x pattern for it. But the V5 didn't have a 6x pattern, so that one is Canadian-made; it would explain a lot
I'd suppose that in most cases, 6x would look better than 4x, but in this one, it doesn't. Or maybe 6x is often worse than 4x, and we better wait for 8x or 9x to have acceptable performance?
Uttar
No windows zoom, nor is it 4x.
It is almost impossible to say which is better. The images are just too small, and both have sufficent samples to produce good AA in this example. The left image has at least 6 samples. The right image has at least 5.
As has been said, the comparison is not very helpful. The left clearly does a better job at removing stair stepping, but the way the image is cropped, I'm not sure if the angle is subtly different from the one on the right, so is that the method or the behaviors of the methods at the slightly differing angles? Without textures, multiple angles, or high contrast in the presence of a bright color on the edge to stress the blending, I'm not sure if the comparison is helpful at all.
Also, the question is "which looks better" not which anti-aliases better, and some people have stated they prefer stair-steps for object edges, since it looks "sharper" to them. I think blowing it up will cause those people to focus on that perception in the larger images whereas I don't think the original image gives an impression of blurriness, i.e. the impression of "blurriness" is a factor at lower resolutions. For that, the one on the right looks "sharper" (blown up atleast).
Also, the question is "which looks better" not which anti-aliases better, and some people have stated they prefer stair-steps for object edges, since it looks "sharper" to them. I think blowing it up will cause those people to focus on that perception in the larger images whereas I don't think the original image gives an impression of blurriness, i.e. the impression of "blurriness" is a factor at lower resolutions. For that, the one on the right looks "sharper" (blown up atleast).
Bigus Dickus
Regular
Hard to tell. Damn jpeg.
Even the number of samples is hard to tell because of that, and it's screwed with the sample values as well. Useless comparison for the most part, so I'll vote for the last option, as that is as close to "irrelevant" an option as there is.
Even the number of samples is hard to tell because of that, and it's screwed with the sample values as well. Useless comparison for the most part, so I'll vote for the last option, as that is as close to "irrelevant" an option as there is.
Hehe lol you are taking it too seriously.
It's just that I posted one pic and another guy posted another one.
I thought one looked better he the other so I just wanted to hear what other people think.
If I post teh full sized pics it will show what modes/cards are used thus I cropped them to just show some edge AA. (Of course they are at a slliiiiightly different angle but not by much)
btw dickus, it's the "12" quality setting in Photoshop can't really blame compresion for anything here.
It's just that I posted one pic and another guy posted another one.
I thought one looked better he the other so I just wanted to hear what other people think.
If I post teh full sized pics it will show what modes/cards are used thus I cropped them to just show some edge AA. (Of course they are at a slliiiiightly different angle but not by much)
btw dickus, it's the "12" quality setting in Photoshop can't really blame compresion for anything here.
The pipe demo is very annoying for that exact reason; it's nearly impossible to take the exact same screenshot as someone else.
If you want to compare perfectly identical screenshots, may I suggest a program I made which does just that?
It takes a full 1024x768 screenshot ( or another resolution, there's a .cfg file to change that ) of the exact same scene, drawing a few horribly aliased triangles on screen.
http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=5566
It also does a moving version of the scene, so that you may see the difference with things like QCX, which are more useful in animated cases.
I haven't worked on it for a while, but a new release which includes a .cfg file to modify the triangles position ( mostly for my own ease ) should be available in the coming days. For feedback, please use that nV News thread
Also, yes, I'm aware this is a lame plug. But I hope some people will learn about the existence of that tool and enjoy it
Uttar
If you want to compare perfectly identical screenshots, may I suggest a program I made which does just that?
It takes a full 1024x768 screenshot ( or another resolution, there's a .cfg file to change that ) of the exact same scene, drawing a few horribly aliased triangles on screen.
http://www.nvnews.net/vbulletin/showthread.php?s=&threadid=5566
It also does a moving version of the scene, so that you may see the difference with things like QCX, which are more useful in animated cases.
I haven't worked on it for a while, but a new release which includes a .cfg file to modify the triangles position ( mostly for my own ease ) should be available in the coming days. For feedback, please use that nV News thread
Also, yes, I'm aware this is a lame plug. But I hope some people will learn about the existence of that tool and enjoy it
Uttar
Hehe yeah I've seen it.
But I didn't post both pics.
At first I didn't intend to do any comparisons I just posted the pic you see on the right: (Which is Ti4600 with 8x FSAA btw)
Then someone else posted a similair pic of Radeon 9700 Pro running with 6x FSAA.
Apart from some strange aliasing in the left side of the pic I think the Ti4600 did a better job at cleaning up the jaggies.
I can't access the full pictures since they were posted on another server (geocities or something) that has exceeded it's bandwidth.
http://www.geocities.com/ewetzlma/r300_6x.png
http://www.geocities.com/ewetzlma/nv25_8x.png
-rightclick and save as
(BTW ignore the blown up pic in the nVidia shot, I increased the size in photoshop which used bicubic filtering, I didn't look at it that closely until I've already posted the pic)
But I didn't post both pics.
At first I didn't intend to do any comparisons I just posted the pic you see on the right: (Which is Ti4600 with 8x FSAA btw)
Then someone else posted a similair pic of Radeon 9700 Pro running with 6x FSAA.
Apart from some strange aliasing in the left side of the pic I think the Ti4600 did a better job at cleaning up the jaggies.
I can't access the full pictures since they were posted on another server (geocities or something) that has exceeded it's bandwidth.
http://www.geocities.com/ewetzlma/r300_6x.png
http://www.geocities.com/ewetzlma/nv25_8x.png
-rightclick and save as
(BTW ignore the blown up pic in the nVidia shot, I increased the size in photoshop which used bicubic filtering, I didn't look at it that closely until I've already posted the pic)
Ante P said:
Eh, 8x FSAA? That? But there's only 5 different shades of grey! Oh, right, it's ordered
Well, then it probably wouldn't have been useful to use my program if you didn't take one of the pics. But if he also used it in the first place, comparing would be been easier.
Just trying to make it more used so comparing AA is more objective, with less opportunities for voluntary biasing. I really should enhance the program, get that website running, put a page explaining the program and e-mail some review sites to ask them if they'd like to use it in their future image quality. But I guess it might take a while to do all that
One really nice thing about the program is also that 4 zipped .TIF images only take 40KB, because the colors are all so similar. So putting the images for download on a reviewer's website is perfectly acceptable.
Anyway, I'll stop lamely advertising my program. For now
Uttar
Bigus Dickus
Regular
.jpg compression is altering the sample values. I can't think of any other reason why two adjacent FSAA pixels would have the same color value, at least not in the shot you provided.
Uttar said:Just trying to make it more used so comparing AA is more objective, with less opportunities for voluntary biasing. I really should enhance the program, get that website running, put a page explaining the program and e-mail some review sites to ask them if they'd like to use it in their future image quality. But I guess it might take a while to do all that
I've been thinking of starting using it (@ www.nordichardware.se)
I usually just use a game that does quicksaves and then use that as a basis for comparisons. Our readers aren't often very content if you post too "sterile" pics when comparing IQ.
