Image Quality Showdown: ATI vs. NVIDIA Fall '03 October 27, 2003 Chris Angelini

Introduction

Page:: ( 1 / 8 )

Image Quality Revisited

This past August we brought you an image quality comparison between ATI’s RADEON 9800 Pro and NVIDIA’s GeForce FX 5900 Ultra. We did so with the knowledge that image quality is rather subjective, and one person’s anti-aliasing of choice is the butt of another’s jokes.

That first piece was well received for the most part, but we read every piece of feedback nonetheless, in email form and on the message boards of other sites. Although the original plan consisted of publishing a follow-up piece within a week or two of the first article, a sudden influx of new products kept the entire staff busy, and consequently, round two of the image quality debate was delayed.

A few things have changed since then. Mainly, ATI released its RADEON 9800 XT and NVIDIA unveiled the GeForce FX 5950, both cards fundamentally faster versions of their predecessors. Further, ATI put the wraps on what its own driver team calls the “most innovative and significant CATALYST release ever,” CATALYST 3.8. NVIDIA followed suit by receiving Microsoft WHQL certification on the long-anticipated Detonator 52.16 driver. Clearly, more than a few variables have changed since the last time we looked past quantifiable performance and into the world of image quality.

Changing for the Better

I doubt that any of my colleagues will argue frame rates tell the whole story when it comes to making a product recommendation. Benchmarking is a perilous business, especially with ATI and NVIDIA slinging accusations back and forth. And so it becomes necessary to delve a bit deeper, to analyze the quality of an image when you, the user, select any combination of anti-aliasing and anisotropic filtering.

In the last piece, we used a limited number of titles and matched modes. For instance, ATI’s 4x anti-aliasing was compared to NVIDIA’s 4x; likewise for the anisotropic filtering settings and combinations of the two. Of those comments that we received, a majority asked for maximized settings in the follow-up. That is, you wanted to see ATI’s most intensive options compared against NVIDIA’s.

So, this time around we’ve used the highest settings available on both competing cards purely for the sake of image quality. We’ve also added some shots from the first game scene in 3D Mark03 to highlight anti-aliasing (another popular request). The NASCAR 2003 Racing Season screenshot is a bit different to satiate those who weren’t satisfied with the first comparison, we’ve zoomed out of the IL2: Forgotten Battles scene for a better anti-aliasing shot, and Brandon recorded a custom demo to showcase Unreal Tournament.

Have the latest round of hardware and software releases improved the state of image quality? Read on to find out…


SIDEBAR: D-Link recently released a new firmware that enables speeds of up to 108Mbps over its 802.11g wireless products. If you have any of those components, check out D-Link’s website for more details.

How We Tested

Page:: ( 2 / 8 )

AMD Athlon 64 FX-51 2.2GHz

ASUS SK8V K8T800 Motherboard

1GB Corsair Registered DDR400 Memory

ATI RADEON 9800 XT 256MB (Catalyst 3.8)
GeForce FX 5950 256MB (Detonator 52.16)

Maxtor 250GB SATA Hard Drive (7200RPM 8MB cache) RAID 0

Windows XP Professional with Service Pack 1

DirectX 9.0b

Desktop resolution 1024x768, 32-bit color, 85Hz refresh

Benchmarks

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


SIDEBAR: If you own an ATI card and haven’t yet downloaded the CATALYST 3.8 suite, you may do so here.

3D Mark03 – Game Test 1

Page:: ( 3 / 8 )

3D Mark03 v.330 – DirectX 9

[image]

Notes

The only immediately perceptible difference between these two shots is that the smoke trailing the fourth bomber in the formation is slightly darker in the ATI image, while the exhaust is a bit whiter in the NVIDIA shot.

[image]

Notes

Slight variations are again apparant in the smoke trails in the shots above. Interestingly, 3D Mark03 recognizes ATI’s ability to run 6x anti-aliasing, but it limits the GeForce FX to 4x, probably because the modes above 4x are combinations of multi-sampling and super-sampling.

Nevertheless, the shots above are both taken using the 4x anti-aliasing methods. And while NVIDIA’s quality seems to have improved, ATI is still clearly dominant. Check out the magnified shots of the rear gun turrets for an example. The difference is still pretty significant.


SIDEBAR: If you own an NVIDIA card and haven’t yet downloaded the Detonator 52.16 suite, you may do so here.

3D Mark03 – Game Test 4

Page:: ( 4 / 8 )

3D Mark03 v.330 – DirectX 9

[image]

Notes

It’s tough to make a call on Game Test 4, even without anti-aliasing or anisotropic filtering enabled. The two images are nearly identical, but there are slight differences between the two. Odd pixels rendered in one are not rendered in the other, and vice versa.

[image]

Notes

Both cards achieve similar anisotropic filtering here, and there is very little difference, once again, between the RADEON 9800 XT and GeForce FX 5950. If anything, the ATI shot is a tinge darker.

[image]

Notes

NVIDIA has made up a lot of ground in its anti-aliasing quality since the last time we explored it. The odd blurring issues we previously encountered are fixed nearly to the point of being comparable. At this point, the RADEON 9800 XT has seemingly better quality on objects in the foreground, while the GeForce FX 5950 is able to display a neater image on the background. For example, the foreground grass is fuller in the ATI shot, evidenced by the blades in the bottom right hand corner. Meanwhile, the NVIDIA shot does a much better job on the tree branches in the background.


SIDEBAR: DVD burning is advancing at an incredible pace – I picked up Plextor’s latest 8x DVD+R writer and completed an entire DVD disc in about nine minutes!

NASCAR Racing 2003 Season

Page:: ( 5 / 8 )

NASCAR Racing 2003 Season – DirectX 8

[image]

Notes

There is nothing notable here, folks. Both cards display very similar images, which includes a headless back row of spectators, a bug that seems to be plaguing both cards.

[image]

Notes

Then, something interesting happens. We cranked the GeForce FX 5950 up to 8x anti-aliasing and the RADEON 9800 XT to 6x anti-aliasing, the maximum settings for both cards. The GeForce FX 5950 image improves dramatically; of course, the jagged edges are gone, but the textures are sharper as well. Meanwhile, the RADEON 9800 XT persists in decapitating the back row of onlookers. And while the jagged edges disappear, the card still retains poor texture quality seen in the preceding image. Consequently, the track is more detailed in the NVIDIA shot. It looks like ATI’s anti-aliasing is still superior to that of NVIDIA’s, but the discrepancies in detail are so distracting that it’s hard to tell.

[image]

Notes

The GeForce FX 5950 supports 8x anisotropic filtering and the RADEON 9800 XT supports up to 16x, so those are the settings we’ve used here. The use of anisotropic filtering has a massive effect on the scene’s overall detail level, but there is still a distinct difference between the competing cards. NVIDIA’s texture clarity is progressively better than the previous scene, though the jagged edges are back due to the lack of anti-aliasing. ATI’s texture quality is much closer to that of the GeForce FX; mainly, it isn’t as bad as the baseline shot. Plus, the spectators have their heads back!

[image]

Notes

In our previous image quality article we argued that 4x anti-aliasing and 8x anisotropic filtering provided a good balance between graphical detail and performance. Although we’re focusing primarily on maximized settings this time around, it’s still important to consider the midrange settings that are more realistic at high resolutions.

With both AA and AF turned on, the overall package is a mixed bag. ATI turns in better anti-aliasing as seen on the concrete barrier in the background and the fence in the foreground. The grass in the background is still noticeably blurrier than the GeForce FX 5950, though.

[image]

Notes

There isn’t much argument here - with all of the visual details maxed out, the GeForce FX 5950 delivers a very nice picture. The difference in anti-aliasing quality is debatable, though you’ll invariably have to visit the topic of playability before settling on a favorite. As before, the background grass is just a bit clearer on the GeForce FX than it is on the RADEON 9800 XT.


SIDEBAR: Does anyone have a recommendation for an affordable cell phone that can be used on AT&T’s network? I’m looking to replace my old Nokia 8265.

IL-2 Sturmovik: Forgotten Battles

Page:: ( 6 / 8 )

IL-2 Sturmovik: Forgotten Battles – OpenGL

[image]

Notes

Both images are very similar, save the darker background in the RADEON 9800 XT picture. As far as quality is concerned, they both do very well.

[image]

Notes

The differences between ATI’s 6x mode and NVIDIA’s 8x anti-aliasing are subtle, at best. That said, the RADEON 9800 XT provides what looks like a smoother image. Of course, then there’s the issue of playability. In IL2, both cards run just fine, though we can imagine other applications where 8x anti-aliasing would simply be unplayable.

[image]

Notes

Again, there isn’t much to separate the ATI and NVIDIA cards with anisotropic filtering maxed out. The background of the ATI card does seem a bit shaper, albeit darker, while the NVIDIA shot blends a little better, with less sharpness that the RADEON 9800 XT.

[image]

Notes

Although we snapped the maximum quality pictures at the same point (at the beginning of this map), the images are somewhat offset. What we end up with is a combination of the aforementioned effects. That is, anti-aliasing looks a bit better on the ATI card, while the image’s background grass is darker. The GeForce FX 5950 looks good in its own regard, though the anti-aliasing quality is slightly below that of ATI’s RADEON 9800 XT.


SIDEBAR: If you enjoy playing Command and Conquer: Generals, you can usually find me online under the handle ‘crazipper.’

Unreal Tournament 2003

Page:: ( 7 / 8 )

Unreal Tournament 2003

[image]

Notes

Even after flipping back and forth between these two shots, it’s tough to discern any differences between the two. If a deviation were to be identified, it’d be that the sky in the RADEON 9800 XT picture is slightly bluer than the GeForce FX 5950. That’s about it, though.

[image]

Notes

Unreal Tournament 2003 lacks many of the hard lines you’d find in a flight simulator. Therefore, even with anti-aliasing maxed out, its really tough to see a difference between ATI’s RADEON 9800 XT and NVIDIA’s GeForce FX 5950. However, what is clear is the difference in image clarity, especially on the floor in the scene we’ve depicted here. If you look closely at the magnified image, the RADEON 9800 XT is not as sharp as the GeForce FX 5950 shot.

[image]

Notes

The nod actually goes to ATI here. Even though both images are quite nearly identical, the GeForce FX 5950 is missing some detail on the arch at the far end of the picture. It’s not a big deal, but we can see from the preceding images that it indeed should be there, and ATI’s RADEON 9800 XT successfully displays the image, as it should.

[image]

Notes

Besides the fact that the RADEON 9800 XT is a shade darker than NVIDIA’s flagship, there is too little difference between the two max quality images to make a definitive judgment. Fortunately, the texture blurring we previously saw on the ATI card isn’t a problem at all here, and for all intents and purposes, we’d call this a tie.

Texture filtering in Unreal Tournament 2003

Back in July, Brandon reported on a bit of controversy surrounding NVIDIA’s texturing method in Unreal Tournament 2003. At the time, it was discovered that both ATI and NVIDIA employ a pseudo-trilinear routine with the respective drivers set for high-quality mode. However, with “Application Preference” selected in the driver, ATI’s cards submit to the game’s settings, as they should, while NVIDIA’s card continues forcing its own performance-maximizing setting.

[image]

Conclusion

Page:: ( 8 / 8 )

There are a few points to take away from our little venture into the world of image quality. Most notably, NVIDIA is in a much better position today than it was several months ago, though that much was probably apparent from our GeForce FX 5950 review.

At the end of the day, we’d still conclude that ATI’s anti-aliasing is superior to NVIDIA’s, a situation that is amplified when you consider game play. There’s no way you’ll be able to run around in Unreal Tournament 2003 at 1600x1200 with 8x anti-aliasing on the GeForce FX 5950. Even 4x AA is pushing it when you’re employing such a high resolution.

That said, in certain games, ATI texture quality has dipped below what we witnessed in our first image quality piece. Whether this is a result of changes made between CATALYST 3.6 and 3.8, we’ll be exploring in the very near future. The issue improves with anisotropic filtering, so we’d recommend anisotropic filtering as a necessity on the RADEON 9800 XT, especially if you’re into NASCAR Racing 2003 Season, which demonstrated the most noticeable blurring. The issue also surfaced in Unreal Tournament 2003 (again with anisotropic filtering disabled). Interestingly, IL2 and 3D Mark03 didn’t exhibit the problematic behavior.

Finally, even though NVIDIA’s image quality in Unreal Tournament is really quite good, it should be said that the firm still hasn’t made good on its promise to fix the “Application” setting in its control panel applet, which would allow Direct3D games to use filtering as intended by the game developer. Instead, it forces a hybrid mode that improves performance and, consequently, benchmark scores.



SIDEBAR: Which is more important to you, anti-aliasing or anisotropic filtering? Were you surprised by the results? Chat with others about this article in the news comments!