Exploring ATI's Image Quality Optimizations [Guru3D.com]

[Guru3D News]

A somewhat heated topic amongst graphics card manufacturers is how to get as much performance out of a graphics card with as little as possible image quality loss. In the past both ATI and NVIDIA have...

More...

[Mkilbride]

So for every ATi review, basically I'll take 10% performance off and that'll be the real performance.

[k3vst3r]

Originally Posted by Mkilbride

So for every ATi review, basically I'll take 10% performance off and that'll be the real performance.

pretty much

[Matt26LFC]

Originally Posted by Mkilbride

So for every ATi review, basically I'll take 10% performance off and that'll be the real performance.

Unless stated in the review that the image quality setting has been changed to HQ. Obviously reviews on this site deduct around 8% as Hilbert i believe said he leaves it on its default setting.

Don't think it bothers me too much that they've done this, i know i can manually change the setting to HQ if i want too. Not that i own a AMD card anyway

[JonasBeckman]

Hasn't it been like this for a really long time though, I don't know much about this but as a older generation ATI card user I don't have the new AI options but it is the same technique as with the Mip-Map Filtering Quality option isn't it and by default that defaults to high and not very high and similarly ATI's defaults are "balanced" instead of "very high" for the default, non-advanced view of the CCC and have been for a long time hasn't it?
(Can't say how it's changed for ATI 5K and 6K series but I imagine the defaults are comparable.)

EDIT: NVIDIA is similar in a way also no? balanced defaults for the non-advanced view though more control over if you want that to decide, override it with the default settings or use a mix of both along with app profile settings which can also apply optimizations.
(Mostly related to trilinear and AF optimizations if the view on the second computer with it's 7800 GTX is still accurate, disabled and grayed out when switched to high quality instead of the default quality option.)

[Ven0m]

The problem arises when we compare 2 cards - one by AMD, one by NV and AMD marginally wins. The first impression for quite a lot of people is that AMD is faster, which is not the case in this scenario.

Because of that, the readers should be explicitly informed that these tests are performed with different image quality settings, as we're not really comparing cards in 100% fair way. Other solutions - decrease quality settings for NV cards (not cool), let NV get worse scores because they care about IQ more (not cool either).

We may say that it's a rare case. It's not really visible in many third-person perspective or strategy games, but in games with low camera angle, it may be annoying. If you play racing games / MMOs / FIFA series, etc, then comparing AMD and NV cards at default settings without IQ notice is just unfair.

[nicugoalkeper]

Sorry to say it but ATI is doing this for speed not quality !
So +1 to Nvidia (higher price but better quality and other nice features)

[cold2010]

http://ht4u.net/reviews/2010/amd_rad...ed/index11.php

[John Dolan]

Ive used half a dozen of each brand over the last decade or so and ive always thought that the ATI cards gave better IQ.The older nvidia 7 series used to look particularly bad in comparison.

[WhiteLightning]

ok its clear, im a cheater for about 2 years now aargh....

[Undying]

Originally Posted by WhiteLightning

ok its clear, im a cheater for about 2 years now aargh....

No one is cheater people just get what they pay for.

[k3vst3r]

HH says recently

Currently with the Radeon HD 6000 series release the Catalyst drivers (Catalyst AI) have a new setting which allows control over control texture filtering with settings for 'High Quality', 'Quality' and 'Performance'.

High Quality turns off all optimizations and lets the software run exactly as it was originally intended to. Quality, which is now the default setting - applies some optimizations that AMD believes remains objective and keeps the integrity of the image quality at high levels while gaining some performance. The last setting is the Performance setting which applies supplementary optimizations to gain even more performance.

What HH is explaining is newest drivers have optimizations enabled by default where as nvidia don't.

[Exodite]

This is getting tiresome.

Both vendors use optimizations that can adversely affect image quality in the default settings. Both vendors allow such optimizations to be disabled.

It's completely inane to apply some kind of scaling pulled entirely from the backside to 'compensate' for this when testing different cards. Obviously testing should be done at default settings and image quality compared.

If there are obvious differences in image quality that has to be accounted for when reviewing a product, obviously, but reading quite a lot of reviews both here and on other sites it's obvious there are no such differences.

Bar the odd bugs, obviously.

[VultureX]

The first comparison shot is not clear to me.

Why does it say NV(16xQ) in the left screenshot (looks like an nvidia only anti-aliasing mode to me) and 16xAF in the right screenshot?

Also the article does not make it clearer:

To your right you can see NVIDIA at work to the left an ATI Radeon 5000 series card

And further down:

You can see that the right image from the Radeon 5000 card

The left screenshot looks better, so I guess that would be Nvidia... But do they use exactly the same AA and AF settings? Otherwise this would not be a fair comparison:s

Another point would be the use of jpg compression... I'd take comparisons and save the screens as lossless png to do it right...

[TDurden]

Difference is visible in some games..AMD on the right.
http://www.3dcenter.org/image/view/3746/_original
http://www.3dcenter.org/image/view/3747/_original
http://www.3dcenter.org/image/view/3731/_original

[wolvy]

And what about the users with lower range video cards?I`m not a big quality freak, i don`t play games alot , but when i do i want them to be playable, not running with 25 fps on ULTRA HIGH settings.Those extra 8% performance gain are more than welcome for me, "cheating" or not.....

[Lane]

I have one question anyway, have you make the test of fps using an older driver and the new one with new settings option ? it look like you use only HQ and perf for determine the difference on the same driver .. or each driver profile increase--- it's same for nvidia if you remove all optimisation on the driver ( optimisation/by game ) you will get a similar down on performance.

it was too the case if you set CAT AI off with old drivers ..( cause there's no more any profile by game and so bug and other fixes, + specific optimisation will not work, including crossfire profile.. )

If you want to compare fps lost or gain, you need test with the Cat 10.9 or 10.10 with the driver set on Cat "standard " and then set with new catalyst " Quality ", HQ etc... and then you can see how much between you gain loose between " quality on new driver " and standard on old.. i really doubt the difference is 8%, most likely 1% ( so at max 1-2fps, nothing who can make you say they cheat for it.)

Trackmania on HD5870 is the worst example, as this game fail with a problem on the AF algorythm who bug with noisy textures, you can't use this game for a comparaison, cause whatever is the the reason, this is not due to a Optimisation or wathever ....

Originally Posted by TDurden

Difference is visible in some games..AMD on the right.
http://www.3dcenter.org/image/view/3746/_original
http://www.3dcenter.org/image/view/3747/_original
http://www.3dcenter.org/image/view/3731/_original

go read the article of 3D center claiming AMD have a hardware problem on HD5870-HD4870-HD3870 with AF, and you will see they have allready used TM ...

In reality this make 3 years 3Dcenter use TM, HL2 for claming the difference in AF between ATI and Nvidia...

This is one concern i have about all of this, why don't use BC2, Dirt2, F12010, COD for check quality difference, and use instead games like HL2 ( and strangely enough not L4D), Quake 3, TM, or Oblivion .... since how many year you have see a review who use thoses games ?

[Hilbert Hagedoorn]

Originally Posted by VultureX

Another point would be the use of jpg compression... I'd take comparisons and save the screens as lossless png to do it right...

Fixed, right is Radeon, obiously.

On the PNG/JPG. The full blown 24-bit PNG or even BMP is 13 MB per image, I opted JPG at maximum quality (100%) . With the screenshots at 2560x1600 pixels you will not notice the difference.

[Darren Hodgson]

When I had my HD 5870 CFX setup, I would always up the (I think they were) texture filtering settings from Quality to High Quality so I was always aware that by not doing that I was compromising the image quality. Now I'm back with NVIDIA I noticed that they too default to a texture filtering setting of Quality and, again, I have raised that to High Quality at the cost of a little performance.

So, unless I'm misunderstanding this article, both NVIDIA and AMD seem to apply the same texture filtering optimisations and neither use the best quality settings, it is up to the end-user to select them.

[k3vst3r]

Originally Posted by Darren Hodgson

When I had my HD 5870 CFX setup, I would always up the (I think they were) texture filtering settings from Quality to High Quality so I was always aware that by not doing that I was compromising the image quality. Now I'm back with NVIDIA I noticed that they too default to a texture filtering setting of Quality and, again, I have raised that to High Quality at the cost of a little performance.

So, unless I'm misunderstanding this article, both NVIDIA and AMD seem to apply the same texture filtering optimisations and neither use the best quality settings, it is up to the end-user to select them.

quality on nvidia doesn't have texture filtering optimizations enabled

What HH is saying is you get 100% image on quality setting for nvidia cards by default but 99% image quality on ATI cards by default with latest drivers

[Lane]

Originally Posted by k3vst3r

quality on nvidia doesn't have texture filtering optimizations enabled

What HH is saying is you get 100% image on quality setting for nvidia cards by default but 99% image quality on ATI cards by default with latest drivers

It's not what saying Nvidia when they respond to it, they say they have thoses optimisation, but they don't impact the Quality. Something you can't only check by compare all beta and driver one by one, and ofc nobody want to do this.

NVIDIA Technical Marketer Jeffrey Yen

I think there's a misunderstanding with how our profiles function. The complete quote in our guide should be "NVIDIA's official driver optimization's policy is never to introduce a performance optimization via .exe detection that alters the application's image quality, however subtle the difference."

That doesn't mean that profiles don't look for .exe files. Just that we're unwilling to alter the application's image quality.

I'm sure you're familiar with many of the performance improvements across games and other applications that our drivers have enabled over the years.

[Moricon]

Is this a big deal..NO!

Should AMD default to High Quality in Driver..NO!

Should AMD publically announce the correct settings for comparable benchmarks for Hardware Testing...YES!

This is a simple solution. AMD should just make consumers aware of the settings in the Drivers to allow people to make their own mind up if they prefer the optimisations or not, and they should inform every hardware reviewer of these settings and advise like for like settings across the different platforms for fair comparison!

They have not done anything wrong with these optimizations, they are good performance boosters for very very little image quality hit, so small its really not noticable. I have only one noticed it in LOTRO, running with the camera view at a certain zoom and angle I get that exact AA effect of a solid gridline, but setting to High Quality in CCC removes this effect, no other game have I seen this happen.

We all know AMD does not have the faster cards, that belongs to Nvidia..but you pay the price for the faster cards! If you want performance for your pound, the way is AMD! If you want pure performance cost irrespective, move with Nvidia!

[Kohlendioxidus]

what's the point of this thread??

I don't play with a microscope conected to my eyes and I don't see the reason of debating if ATI or nvidia should use "Quality" or "High Quality" settings...It's up to user to decide. ATI always had more IQ expecially when watching movies. Regarding games I see NILL difference...maybe they are but at microscopic scale...

[chanw4]

How did people come to the conclusion that Nvidia did not use optimization on default setting and that you need to deduct 8% in performance to get 'real performance' or raw perfomance?

[alanm]

Originally Posted by Kohlendioxidus

what's the point of this thread??

I don't play with a microscope conected to my eyes and I don't see the reason of debating if ATI or nvidia should use "Quality" or "High Quality" settings...It's up to user to decide. ATI always had more IQ expecially when watching movies. Regarding games I see NILL difference...maybe they are but at microscopic scale...

The 'point of this thread' is that unless this is addressed by ATI - and only in regards with what they supposedly did with these 2 drivers (10.10 and 10.11?) default settings and if they continue it - then it may hang over their head with continued controversy. I can just see future bench comparisons with Nvidia owners saying 'uh, but you have to deduct 8% off the ATI figures to make it fair', etc.