PDA

View Full Version : fx's AF quality


heedory
07-04-03, 12:03 PM
http://home.megapass.co.kr/~heedory2/nv8.jpg
http://home.megapass.co.kr/~heedory2/r8.jpg

better than 9800..!!
I've never known it before..!!
why every guys say that radeon has better image quality?
this pic clearly show that's not true

The Baron
07-04-03, 12:09 PM
A. those things are way too compressed for an actual IQ comparison
B. yes, the NV cards have better AF because it is applied to every surface regardless of the angle, while ATI cards apply AF differently depending on the angle of the surface (or something like that, I forget). but, ATI has superior AA algorithms.

Hanners
07-04-03, 01:14 PM
Originally posted by The Baron
A. those things are way too compressed for an actual IQ comparison
B. yes, the NV cards have better AF because it is applied to every surface regardless of the angle, while ATI cards apply AF differently depending on the angle of the surface (or something like that, I forget). but, ATI has superior AA algorithms.

Actually, I'd move on a bit further than that to say that the 5900 has better AF than the R3x0 cards, but they in turn have better AF than the rest of the GeForceFX line-up.

I agree though, those pics don't show anything.

The Baron
07-04-03, 01:26 PM
I've always heard that the 5900 has the same AF and AA implementations as every FX card... but you're saying it doesn't?

B&R
07-04-03, 04:39 PM
wat exactly are the AF settings?/

euan
07-04-03, 05:10 PM
<fanatic>No way that GF-FX picture willl look hideous when seen in motion, as there will be way to much shimmering from the pixels. Looks like no AF with crap textures to me!</fanatic>

lol :D

All AF implementations have downsides.

NV's is the performance hit.
Ati's is the obscure angles. It's changes depending on the z-roll.

Hanners
07-04-03, 06:17 PM
Originally posted by The Baron
I've always heard that the 5900 has the same AF and AA implementations as every FX card... but you're saying it doesn't?

A lot of the reviews seemed to make a big thing about how aniso was improved with NV35, and finally on a par if not better than the R3x0. I know part of it was down to drivers, but I got the impression some of the improvement was also in silicon.

StealthHawk
07-04-03, 07:27 PM
Originally posted by Hanners
A lot of the reviews seemed to make a big thing about how aniso was improved with NV35, and finally on a par if not better than the R3x0. I know part of it was down to drivers, but I got the impression some of the improvement was also in silicon.

That was a driver change. It first happened in 43.51, not in 44.03 as reviwers claim. It was a retroactive change for all gfFX cards.

StealthHawk
07-04-03, 07:28 PM
Originally posted by euan
All AF implementations have downsides.

NV's is the performance hit.
Ati's is the obscure angles. It's changes depending on the z-roll.

nvidia's AF has around the same performance hit as ATI's AF...maybe lower.

Of course, we may have to take that with a grain of salt, since nvidia has been known to do some fishy things with their AF. But also, ATI's Quality AF isn't always doing trilinear either :p

Nv40
07-04-03, 07:51 PM
Originally posted by heedory
http://home.megapass.co.kr/~heedory2/nv8.jpg
http://home.megapass.co.kr/~heedory2/r8.jpg

better than 9800..!!
I've never known it before..!!
why every guys say that radeon has better image quality?
this pic clearly show that's not true


the irony is ,that i dont see the same sites ,that -nitpick- IQ diferences in the Geforce5900 recently in unrealT :rolleyes: ,doing the same things with ANisotropic filtering "optimizations",where ATI is clearly is taking free pass and free performance a the expense of IQ.
by how much? to the point where PErformance AF mode 8x in Nvidia cards
looks better than ATI 8x/16 HQ in many angles .
i dont need the yellow circles to notice here the diferences.
that why Digilife test ATI in 16x AF and NVidia cards in 8xAF. :)
that same FRENCH site who posted the screenshots at least made aware -many times in their reviews- that ATI AF is cutting corners at the expense of IQ.the more reviewers do this ,the more will force ATI to stop their agressive optimizations. you know there are many more games than deathmatch FPS that plays mostly in flat surfaces. :rolleyes:

i have no problems with extensive IQ investigations from ATI or NVidia cards, how they are doing things. the more people investigate ,the more IHV will work to improve IQ. the problems are when The same people only picks X company , but "ignore" completely what Y company is doing , because obviously it will not made look their favorite company in the best light. i dont see enough investigations about this. because some people are aware of this diferences ,that means is ok to give free pass to a company? if NVidia "cut corners" or optimize to much to gain performance is BAd ,but if ATI do it ,then is a valid optimization . :rolleyes:

extreme_dB
07-04-03, 08:45 PM
Why do all sites compare Nvidia's 4xAA with ATI's 4xAA in benchmarks? ATI's 16xAF should be compared with Nvidia's 8xAF, but Nvidia's 4xAA should be compared with ATI's 2xAA.

Nv40
07-04-03, 10:35 PM
Originally posted by extreme_dB
Why do all sites compare Nvidia's 4xAA with ATI's 4xAA in benchmarks? ATI's 16xAF should be compared with Nvidia's 8xAF, but Nvidia's 4xAA should be compared with ATI's 2xAA.

good point.
maybe because we cant live without comparisons..? :)

when it comes to AA , 4xAA in ATI is not comparable to Nvidia 4xAA.
they are diferent techniques,and by the screenshots posted looks that ATI have a more efficient algoritm at 4x and 6x.. i can accept that. the most comparable way with ATI and NVidia is at 2x ,because use the most similar techniques.however is not 100% perfect comparison. thats impossible ,not in the past ,not today and never in the future.because there are no Rules set in stone of the "right way" AA or AF should work. NVidia 4xS looks very nice ,but takes more peformance ,because is using a mix between diferent techniques .
its up to the reviewers to choose wich modes ,they will use..
what it is really important is that web reviewers (like the FRENCH web site) clearly inform the public how both companies AA and AF work.and why they will compare X settings vs Y setting. which one works better and and which one is faster but give less IQ .

what i cant understand is when WEbsites reviewers Nitpick too much and complain ,by how "wrong" is the techniques of X company.but at the same time they just "ignore" and never see anything "wrong" about what Y company is doing. :rolleyes: unfortunately ,comparisons againts diferent companies are always necessary (or maybe not - you still can compare againts video cards of the same company). however to be impartial in a review,with your conclusions is something very rare to see these days.
Thats why my best advice to people ,is to forget the opinions of reviewers and buy only the best video cards -for the games you are going to play.
because as we already know there are games that runs faster in ATi cards and others in Nvidia cards. the "best" VIdeo card doesnt exist anymore.
or the absolute winner. only exist what works better for your needs .

-=DVS=-
07-04-03, 11:02 PM
Originally posted by heedory
better than 9800..!!
I've never known it before..!!
why every guys say that radeon has better image quality?
this pic clearly show that's not true [/B]

Big deal found one game where ATI looks worse then Nvidia :rolleyes: how about checking up all games ever made and then compare how much of them look better on Nvidia and how many on ATI then we can declare a winner :p

Fact is ATI does have worse AF on some angles , but its not like every game runs on those angles ;)

Fact ati have supperior AA to Nvidia

Witch card would you like a card with inferior AF on angles compared to competition but Awesome AA (on speed & quality)

Or card with perfect AF(sharp AF on all angles) but poor AA (quality & speed wise)

There aren't many games that ATIs AF looks bad but AA looks nearly always way better.
Offcourse if you only play those games that happen to look better on GFFX only , its obvious choice.

Correct me if im wrong :p

Behemoth
07-05-03, 03:41 AM
Originally posted by extreme_dB
Why do all sites compare Nvidia's 4xAA with ATI's 4xAA in benchmarks? ATI's 16xAF should be compared with Nvidia's 8xAF, but Nvidia's 4xAA should be compared with ATI's 2xAA.
yeah i always think both companies' AA and AF comparisons are not 100% apple to apple, you need to balance things out by putting their AA/AF pros and cons into consideration. looking only at the numbers, e.g. 4x vs 4x, scores vs scores, is quite misleading. ati's AF do quite badly at worst angles, nvidia's AA do very badly on near horizontal and vertical edges but ati's AF beat nvidia's at around 0 and 90 degrees whereas nvidia's AA look worse than ati's at just about any angle.

Behemoth
07-05-03, 04:08 AM
Originally posted by -=DVS=-
Big deal found one game where ATI looks worse then Nvidia :rolleyes: how about checking up all games ever made and then compare how much of them look better on Nvidia and how many on ATI then we can declare a winner :p

Fact is ATI does have worse AF on some angles , but its not like every game runs on those angles ;)

Fact ati have supperior AA to Nvidia

Witch card would you like a card with inferior AF on angles compared to competition but Awesome AA (on speed & quality)

Or card with perfect AF(sharp AF on all angles) but poor AA (quality & speed wise)

There aren't many games that ATIs AF looks bad but AA looks nearly always way better.
Offcourse if you only play those games that happen to look better on GFFX only , its obvious choice.

Correct me if im wrong :p
its not about this one game, all games would reproduce this problem faithfully on those funny angles. its just not as ugly as jaggies.
people claim 9800pro has superior IQ is largely due to their superior AA, if you turn off the AA and compare AF only, the answer is not as clear cut.

DaveBaumann
07-05-03, 04:12 AM
Originally posted by Nv40
the irony is ,that i dont see the same sites ,that -nitpick- IQ diferences in the Geforce5900 recently in unrealT :rolleyes: ,doing the same things with ANisotropic filtering "optimizations",where ATI is clearly is taking free pass and free performance a the expense of IQ.
by how much? to the point where PErformance AF mode 8x in Nvidia cards
looks better than ATI 8x/16 HQ in many angles .
i dont need the yellow circles to notice here the diferences.
that why Digilife test ATI in 16x AF and NVidia cards in 8xAF. :)
that same FRENCH site who posted the screenshots at least made aware -many times in their reviews- that ATI AF is cutting corners at the expense of IQ.the more reviewers do this ,the more will force ATI to stop their agressive optimizations. you know there are many more games than deathmatch FPS that plays mostly in flat surfaces. :rolleyes:

We make the limitations clear in each review major architectural review we do - the fact is, the limiation of ATI's hardware are known, and documented. Should that change we'll talk about it.

The situation with UT2003 is different since NVIDIA are promoting reviewers to to display the what type of filtering is given with the quality settings, but where UT2003 is concerned they are not even giving you the option of having that - this was not documented until now.

scott123
07-05-03, 07:45 AM
I have both the 9800 Pro and the 5900 Ultra. I play Nascar 2003 like crazy. The 9800 Pro running in D3D has MUCH better FSAA, but is not quite as good in the Anisotropic filtering department. However the difference between FSAA on the 2 cards is much greater then the anisotropic filtering. Nascar 2003 is a very CPU intensive game, but in every test I ran (in D3D), the 9800 Pro beat the 5900 Ultra in FPS (by a decent margin). When running Nascar 2003 in OpenGL, the tables are turned, and the 5900 Ultra wins (by a decent margin also). Nascar 2003 looks best in D3D, so overall, the nod goes to the 9800 Pro, if this is a game you play a lot. This game is a good one for showing the differences in FSAA between the two cards, due to the vast distances that are drawn.

Scott

Rogozhin
07-05-03, 04:19 PM
The main difference between ati's and nvidia's AF algo seems to be the sample pattern shape, with ATI using a rectangular pattern, and nVidia using a four-sided polygonal pattern that will change depending upon the degree of slope-related distortion along the x and y-axes. nVidia can vary the number of samples and the sample pattern "footprint," whereas ATI can vary only the number of samples. Radeon 9700's sample pattern is always rectangular, which in many instances poses no problem.

rogo

Nv40
07-05-03, 04:53 PM
Originally posted by Rogozhin
The main difference between ati's and nvidia's AF algo seems to be the sample pattern shape, with ATI using a rectangular pattern, and nVidia using a four-sided polygonal pattern that will change depending upon the degree of slope-related distortion along the x and y-axes. nVidia can vary the number of samples and the sample pattern "footprint," whereas ATI can vary only the number of samples. Radeon 9700's sample pattern is always rectangular, which in many instances poses no problem.

rogo


no problems for the gamers ,but yes for benchmarking .no problems only in the cases where the reviewers post only FLat surfaces screenshots.it can be a real problem for benchmarking ,this is more pronounced with ATI lower end cards ,9000,9100(8500),9200 ,-that still today- ,use the same old radeon8500 techniques (http://firingsquad.gamers.com/news/newsarticle.asp?searchid=4191) not enough time to fix last year old radeon bug? and many web sites that are not aware of this ,test those cards againts Nvidia lower end cards ,which doesnt have those problems.
because the ATi lowerd ends ones not only doesnt like non-horizontal angles but also drops to bilinear filtering taking too much Free performance at the expense of IQ.
i have seen benchmarks were a radeon 8500 manage to outscore a R300 in AF with AA off. in similar AF settings. :rolleyes:
the official ATI word about their latest R300. is that it is an "improved technique" that they only apply AF when "they believe is necessary".
so yes ,is not a problem for gamers ,but it can be a problem in apples vs apples Benchmarks . if one company cut too much corners to increase their performance.

Hanners
07-05-03, 05:41 PM
Originally posted by Nv40
i have seen benchmarks were a radeon 8500 manage to outscore a R300 in AF with AA off. in similar AF settings.

:eek2:

Where did you see that, any links? I can't see that could ever happen, even with the R2x0 using bilinear to the R3x0's trilinear.

I also wish you'd stop referring to R2x0 only using bilinear with aniso as a bug as well - It was a design decision which ATi consciously made whether people like it or not. Yes, it is rubbish compared to using AF with trilinear, but at the time it did a good job of making AF a usable feature.

Nv40
07-05-03, 06:15 PM
Originally posted by Hanners
:eek2:
Where did you see that, any links? I can't see that could ever happen, even with the R2x0 using bilinear to the R3x0's trilinear.


yes that was my face ,when i saw the review.. :eek2:

the diference is minimal ,but still shows how "Agreesive" ATI can go with their AF. i have not recorded the link ,so this is the only one i have found.

http://firingsquad.gamers.com/hardware/radeon9700/page15.asp

im not 100% sure ,but i think i saw another benchmark where the Radeon8500 scored a bit better the R300 even at 1024x768 in a game.




I also wish you'd stop referring to R2x0 only using bilinear with aniso as a bug as well - It was a design decision which ATi consciously made whether people like it or not. Yes, it is rubbish compared to using AF with trilinear, but at the time it did a good job of making AF a usable feature.


i would like to not mention the R2xx cards ,but it is RElevant to do it.

1)because ATI still sells those video cards ,their AF "techniques",still are present in their latest R9xx lowend cards. and still today i can find reviews where those cards tested in AF againts Nvidia cards. :rolleyes:
2)because it is good for reference to see why ATi latest AF in R3xx cards are implemented in the way they are. because is a easy way to gain performance ,without being noticed to do less work.since most REviewers only post Flat surfaces screenshots.however thanks to XMas tools or ANiso tunnels ,more and more reviewers are aware of what ATI and Nvidia are doing with AF . more aware that angle surfaces are important too in our games.

Behemoth
07-06-03, 03:09 AM
Originally posted by Rogozhin
The main difference between ati's and nvidia's AF algo seems to be the sample pattern shape, with ATI using a rectangular pattern, and nVidia using a four-sided polygonal pattern that will change depending upon the degree of slope-related distortion along the x and y-axes. nVidia can vary the number of samples and the sample pattern "footprint," whereas ATI can vary only the number of samples. Radeon 9700's sample pattern is always rectangular, which in many instances poses no problem.

rogo
the BIGGEST difference is, regardless of the underlying sampling mechanism, ati ADAPTIVELY apply full degree of AF on certain angles which results in much LOWER degree of AF on many other angles VISUALLY, and thats all that matters.