Image Quality Showdown: ATI vs. NVIDIA August 21, 2003 Chris Angelini

Introduction

Page:: ( 1 / 7 )

The Fairy Tale

Once upon a time, evaluating graphics hardware was a simple, straightforward process. You could get away with a couple of benchmarks, a bit about the card itself and a conclusion that compared the product in question to its competitors.

Or maybe it just seemed that way, since there were far fewer features on which to focus. Nowadays, you have to consider 2D quality, video capabilities, and 3D performance. And of course, within each category there is a subset of brand new enhancements to be covered. Then, there’s the need to filter marketing terminology, much of which is useless blather anyway. In fact, next to sound cards, graphics products are perhaps the most difficult to judge. There are a number of variables to consider and admittedly, many of them are difficult to critique without intimate architectural knowledge. For example, who’s to say that a given card’s effective bandwidth specifications are accurate? How efficient is HyperZ? What about NVIDIA’s Lightspeed Memory Architecture? You get the idea…

There are, however, two fundamental criteria that can be used for measuring the value of a graphics card. It doesn’t matter how advanced they get; it doesn’t matter how expensive they become; it doesn’t matter if the next generation of graphics processors is comprised of 300 million transistors, 16 pixel pipelines, and DirectX 10 compliance. At the end of the day, performance and image quality are the two most important factors. Further, each and every hardware enthusiast is capable of passing their own judgment over those two categories. So our job here at FiringSquad, at least as it relates to reviewing a card’s 3D capabilities, is leveraging the experience that comes from testing hardware on a daily basis and playing the latest games to give you an extra bit of ammunition for making an informed hardware purchase.

Why Image Quality? Why Now?

Because performance can be measured quantitatively, it isn’t a secret that benchmark results generally predominate in a review. Hardware vendors consequently optimize their drivers for higher frame rates, knowing that their products will be evaluated on those premises. Unfortunately, abuses occasionally occur on both sides of the fence. Back in 1998, ATI released its Turbo driver to augment performance of the Rage Pro. The only application to benefit, though, was Winbench 98, a synthetic measure of ability. Most recently, NVIDIA added extra clip planes to avoid drawing certain parts of the sky in 3D Mark03’s Test 4. Other questionable optimizations also emerged, some in 3D Mark03 and others in real games.

The end result is that, even while both NVIDIA and ATI have purportedly made policy changes regarding the nature of optimizations, it is important (now more than ever) to take a step further and analyze the quality implications of each company’s respective driver controls.




SIDEBAR: The ALL IN WONDER 9600 Pro should be close to shipping. We can’t wait for IR Blaster support, either.

How We Tested

Page:: ( 2 / 7 )

Intel Pentium 4 2.53GHz

EPOX 4PEA+ 845PE Motherboard

1GB Corsair PC3200 DDR Memory

ATI RADEON 9800 Pro 128MB (Catalyst 3.6)
GeForce FX 5900 Ultra 256MB (Detonator 45.23)

146GB Seagate Cheetah SCSI Ultra320 10000RPM HDD

Windows XP Professional with Service Pack 1

DirectX 9.0b

Desktop resolution 1024x768, 32-bit color, 75Hz refresh

Benchmarks

IL-2 Sturmovik: Forgotten Battles
NASCAR Racing 2003 Season
3D Mark03 v.330


SIDEBAR: Did you know NVIDIA has once again joined the Futuremark development team for future benchmarks? Check out the press release here.

3D Mark03

Page:: ( 3 / 7 )

3D Mark03 v.330 – DirectX 9

[image]

Notes

There isn’t really a discernable difference between the ATI and NVIDIA screenshots at 1024x768, without any visual details enabled.

EDIT 8/22/03: Taking a closer look at the image, you can see slight differences between the RADEON 9800 PRO and the GeForce FX 5900 Ultra, specifically if you look at the white rock on the left, just above the shoreline. Textures on the ATI card are slightly crisper than the GeForce FX, giving the 5900 Ultra a slightly blurrier look in comparison.

Anti-Aliasing

[image]

Notes

There is a much more pronounced discrepancy with 4x antialiasing enabled. The RADEON 9800 Pro maintains sharp textures and smoother lines, while the GeForce FX 5900 Ultra applies a more washed-out effect to the image. To illustrate the difference, check out the rock in the blown-up image (200 percent):

RADEON 9800 PRO


GeForce FX 5900 Ultra

Anisotropic Filtering

[image]

Notes

The anisotropic filtering shots are not very conclusive. Both images are of comparable quality ; even blown up to 200 percent, the variance between them is too subtle to note.


SIDEBAR: NVIDIA published a list of performance and compatibility changes with its 45.23 driver, which is available here.

NASCAR Racing 2003 Season

Page:: ( 4 / 7 )

NASCAR Racing 2003 Season – DirectX 8

[image]

ATI without AA/AF


NVIDIA without AA/AF

Notes

You don’t even need to magnify the images in NASCAR to see how ATI and NVIDIA differ in their implementations. Look down the track in the selected images – the upside-down ZMAX logo (alongside the outer pit wall below the start/finish line) is actually legible in the 9800 Pro screenshot, while it is a little more ambiguous in the 5900 Ultra picture. The further you travel, the more distorted the images become, too.

Anti-Aliasing

[image]

ATI 4xAA applied


NVIDIA 4xAA applied

Notes

Adding anti-aliasing has an interesting effect on the 9800 Pro – it throws the logos that line the track out of focus, a la 5900 Ultra. Yet, ATI continues to demonstrate superior anti-aliasing, as seen in the yellow line above the logos. (Look near the 76 logo coming out of Turn 4 as well.)

Keep in mind that these images here are not magnified or altered. This is exactly how you’d seem them during game play. Of course, they won’t be stills, as seen here but when it comes time to play NASCAR, its anti-aliasing effects are amplified under the influence of motion.

Anisotropic Filtering

[image]

Notes

Again, we don’t need a cross-section to see the winner here. And surprisingly, with 8x anisotropic filtering enabled, the GeForceFX 5900 Ultra regains its lost clarity and bests ATI’s offering. Pay particular attention to the crowd and the stands at the end of the track (above Turn 4), both of which looks better on NVIDIA’s card.

4x Anti-Aliasing/8x Anisotropic Filtering

[image]

Notes

It’s a close race with all of the “eye-candy” turned up. NVIDIA continues to maintain a clear filtering advantage, while ATI’s anti-aliasing implementation is undoubtedly better. This one would have to go to subjective preference, as there are trade-offs apparent in each architecture.


SIDEBAR: If you’re looking for ATI’s latest CATALYST driver package for Windows XP, try looking here.

IL-2 Sturmovik: Forgotten Battles

Page:: ( 5 / 7 )

IL-2 Sturmovik: Forgotten Battles – DirectX 8.1

[image]

Notes

There’s very little difference to be seen between the RADEON 9800 Pro and GeForce FX 5900 Ultra, at least without any visual enhancement.

Anti-Aliasing

[image]

RADEON 9800 PRO 4xAA


GeForce FX 5900 Ultra 4xAA

Notes

As we’ve seen previously, ATI’s 4x anti-aliasing is superior to NVIDIA’s (take a look at the airplane’s wing). But because of the scene’s low fidelity, there isn’t a glaring difference.

Anisotropic Filtering

[image]

RADEON 9800 PRO 8xAF


GeForce FX 5900 Ultra 8xAF

Notes

As seen in NASCAR, NVIDIA holds an edge in anisotropic filtering. As the grass fades away, the GeForce FX screenshot is sharper, while the RADEON 9800 Pro tends to blur a bit.

4x Anti-Aliasing/8x Anisotropic Filtering

[image]

Notes

The same issues that surfaced in NASCAR crop up once again in IL2. The RADEON 9800 Pro does a better job with anti-aliasing, while the GeForce FX demonstrates superior anisotropic filtering.


SIDEBAR: Care for the latest Detonator drivers? Here they are.

UT2003, ATI and NVIDIA

Page:: ( 6 / 7 )

Unreal Tournament 2003

Of course, Unreal Tournament 2003 is one of the real games that has raised eyebrows lately. As Brandon mentioned in his ASUS V9950 Ultra review, both ATI’s RADEON 9800 Pro and NVIDIA’s GeForce FX 5900 Ultra perform a mixed filtering routine when manually configured in the driver. However, when both cards are told to let the application dictate proper rendering, the GeForce FX refuses to abdicate control. The issue isn’t necessarily with specially-optimized drivers per se; the real problem is that NVIDIA is forcing its own display settings, something we don’t care for, especially after spending in excess of $400 on a graphics card. It was previously speculated that the 45.23 drivers would rectify the situation, and while they haven’t, NVIDIA’s general manager of desktop graphics processors, Tony Tomasi, assured us that a fix is in the works and will be included with an upcoming driver set.

ATI and NVIDIA Chime In

In light of the controversy and rumors that have propagated over the past few weeks, we took the opportunity to sit down with both ATI and NVIDIA and get their sides of the story.

Though ATI has been guilty of the same aggressive optimizations NVIDIA has recently taken flak for, its current position seems to be one of concern. Chris Evenden, ATI’s director of marketing, is quick to point out that hand-optimizing programmable shaders for improved performance in one application or another is a slippery slope. Not only does it open the door for fudging image quality, but it also takes time from the driver guys that could be better purposed for making holistic improvements. And once DirectX 9 games begin to proliferate, picking applications for the special “hand optimization” treatment will consist of rounded up all of the applications used to quantify performance. What about the other games, though? Many users may find that their “popular” games run particularly well, while others don’t. In such an environment, performance numbers gleaned from games (and especially synthetic tests) will mean less than ever. Perhaps that is why ATI committed to removing application detection code from its CATALYST driver when it was discovered that 3D Mark03 was special-cased. Chris’s statement, in full:

“Not doing what an application asks is a risky path to start down. It becomes a question of where you draw the line. Economize on shader accuracy here, only apply anisotropic filtering there, and pretty soon you're not giving the gamers the experiences that the developer wanted them to have. The software developers and artists already know that there is a trade off between image quality and frames per second - and they made those decisions as they put the game together. We really see our role to do exactly what they ask, and do it faster than anyone else.”


SIDEBAR: Check out the thread on Beyond3D that exposed the filtering problems in Unreal Tournament 2003 right here.

NVIDIA and Conclusion

Page:: ( 7 / 7 )

NVIDIA Speaks Up

It sounds like ATI worries that optimizations in software (which may or may not impact image quality), may lead to shortcuts in hardware that inevitably damage the viability of the popular APIs. Then again, given NVIDIA’s public stance on driver optimizations, it will be interesting to see if ATI is able to stand firm on its higher ground and continue to put down competitive benchmark numbers. We sure hope so.

On the other hand, NVIDIA’s Tony Tomasi is more optimistic about his company’s ability to make valid optimizations by not affecting image quality. According to Tomasi, software engineers were previously able to make judgment calls on their own optimizations, which is why the 3D Mark03 snafu went down. Through its new policy, however, driver optimizations have to be registered, published internally, and verified by a quality assurance team. Unfortunately, because optimizations are to remain confidential, the only way to uncover incompatibilities between optimizations and other applications is for NVIDIA to catch them, or for the end users to report them. If NVIDIA’s new initiative facilitates better performance and focuses more attention on image quality, then the policy will certainly prove a boon. Then again, it requires more effort than ever from impartial reviewers to ensure quality isn’t taking a back seat to simply winning benchmarks.

Hand-optimized code is only a part of what NVIDIA hopes to accomplish. After all, Tomasi points out that NVIDIA helps every willing developer optimize its code paths. It is also dedicating more resources than ever to developing a run-time compiler for efficient code execution. “Optimizations aren’t new,” Tomasi quips, “shaders simply provide an extra level of programmability with which to work.” It’s understandable NVIDIA would want its shaders to run as quickly as possible, we just hope it can do so without any other adverse effects.

Conclusion

So there you have it, in the tests we've shown you today, ATI tops NVIDIA in anti-aliasing quality while NVIDIA bests ATI in anisotropic filtering. While we only tested each company's flagship, keep in mind that their mainstream offerings utilize the same underlying rendering methods so the results we've just shown you also apply to the GeForce FX 5600 family as well as the RADEON 9600 series. Another point to keep in mind is that these are just static screenshots. Those jagged edges you see on the track in NASCAR often look even worse in motion, making you wonder how you played without AA for so long.

ATI and NVIDIA are currently locked in a pitched battle for 3D graphics supremacy. Each side has its positives and negatives. Deciding which card is right for you will come down to price and availability, performance, 2D and 3D visual quality, and finally, stability/compatibility. Good luck in your search, hopefully this article came a little bit closer to helping you make the decision that's right for you!

SIDEBAR: Which is more important, AA or AF? Voice your thoughts on this article and the graphics market in general in the news comments!