Tech-Hounds.com

The Radeon X800XL and GeForce 6800GT Compared

PCI Express x16 graphics cards
Chipset Radeon X800XL 256 MB, 256 bit, GeForce 6800GT 256 MB, 256 bit

To get a better picture of what these cards have to offer, we've decided to re-evaluate our suite of benchmarks. For one, we decided to update our image test quality comparisons. We use a new scene for Call of Duty and SW: Knights of the Old Republic that will best show how good (or bad) each card deal with AA and AF. That's when we noticed something is not quite right with our Call of Duty scores. We've also re-ran our Battlefield 2 benchmark for the GeForce 6800GTs because the results was, in effect, all over the place.

Call of Duty problems

When we ran FRAPS with the Dawnville demo included in the 1.5 patch, we noticed a very large discrepancy between the results from the timedemo command and what we're getting from FRAPS. Further testing revealed even the on screen fps counter from both reports different frame rates. This happens on both ATI and NVIDIA cards, so we suspect either the in game tool or FRAPS is at fault and reporting inflated scores. We ran every timedemo we have (or can get our hands on). the results are still to far apart. To eliminate FRAPS, we ran it with several games that also feature in-game benchmarking tools, like Splinter Cell, Battlefield 2, Doom 3 and Dungeon Siege. In these games, FRAPS results are either the same or very close, at the very least within 1 fps of each other. With this evidence, we're pretty sure that the in game benchmarking tool in Call of Duty is the one providing the erroneous results. We feel pretty bad about this, since we should've caught this one earlier. This does invalidate our past Call of Duty benchmark numbers, since benchmark results between the timedemo and FRAPS is very different by a significant amount. From now on, we will use FRAPS instead of the in game tool to benchmark Call of Duty.

Battlefield 2

We've re ran and made corrections to our benchmark results for this game in our GeForce 6800GT round up, which you can check here. The cause of the problem is bots - Battlefield is pretty system dependent - if there's too few bots, you'll more likely to have higher frame rates. Thus, fluctuating scores. Now that We've discovered and fixed the problem, performance across the board is more consistent. However, this does not change our conclusion - we can see that performance is relatively the same between all boards.

Showing artifacts in motion and static images

We once said that there are several reasons people buy faster graphics cards: to get higher frame rates, to play at higher resolutions and to play with AA and AF enabled, with full or maximum details. Of course, gamers who like to play with AA and AF will most likely be more concerned with image quality than most people. Higher frame rates are easily to show, but image quality differences are not, because image quality is generally, very subjective. There are some guidelines, but in general image quality comparisons can be very tricky to do. Games may have different rendering paths for each vendor (ATI and NVIDIA) or even implement different levels of detail. Cards from both have different approaches to 3D rendering. Optimizations (valid and not) only complicates the matter further.

The most obvious way of course is to show screenshots taken in game. By flipping through the same screenshot from both cards, we can quickly see if there's any differences. However, keep in mind that any differences present are actually quite normal, because these cards take different approaches (and make different compromises). Blurriness, sharpness, can easily be seen on a static image, but they are less noticeable in motion. Even more so when you're actually playing the game. That's why we categorize any differences between normal differences to noticeable differences.

However, static images often don't really capture aliasing or artifacts (such as shimmering) across different frames. The only way to do that is by showing it in real life on a monitor or a video. Of course, we opted to choose the video - we experimented with different screen capture software that will capture the artifacts we want to show and works 'good enough' with both cards. Unfortunately, that's not easy. We've finally settled on Hypercam, this software lets you screen capture your desktop (or a game) into an uncompressed, 32 bit RGB AVI file. Out of the games we used, only Homeworld 2 and SW: Knights of the Old Republic provide satisfactory results. This way we can provide you a glimpse (albeit in motion) of what we see during testing. In fact, these two games are perfect for this purpose (which we will get into later).

The second problem is showing you the file. Since the recording is a 1:1 pixel record in the form of an uncompressed 32 bit RGB AVI file, it takes lots of space (and bandwidth). A 5 second recording can take as much space as 50 MB. Compressing the video is not an option, since it introduced too much artifacts. After much trial and error, we've decided against using a video, instead making it available as a slide show. We actually extract each frame, in sequence, turn them into BMPs and put them inside a ZIP file that you can download. To view them, you can use a viewer (Windows Picture and Fax Viewer is actually perfect for this purpose), then just press and hold the right cursor button on the keyboard to advance to the next frame. It's just like moving the slider in your video player (frame by frame). However, we did have to made another compromise. The sequence from SW: Knights of The Old Republic was still too large, so we had to compress the BMPs into JPEGs. While there are compression artifacts, much of the aliasing (and lack of) is still very noticeable.

Overview

Here are the technical specification from both manufacturer's website.

ATI X800XL 256 MB
Core Clock : 400 MHz
Memory Clock : 500 MHz (1000 MHz DDR3)
Memory Bus Width : 256 bit

NVIDIA GeForce 6800GT 256 MB DDR3
Core Clock : 390 MHz
Memory Clock : 1000 MHz (500 MHz DDR3)
Memory Bus Width : 256 bit

Image Quality Comparison

A Little Background

Of course, when we're talking image quality in relation to graphics cards, we're talking about image quality with AA and AF. Someone once said, aliasing (or any artifact for that matter) is what you get from trying to do everything too fast. This means you made several shortcuts and compromises to get a high enough frame rate. Anti aliasing (AA) and anisotropic filtering (AF) is a 'hack' to remove these artifacts. From this simple definition, we can base our image quality comparison by seeing which card has the less aliasing.

Let's look at how anti aliasing is done on the GeForce 6800GT and Radeon X800XL. While their implementation is different in some ways, they're actually quite similar. Both cards rely on multisampling anti aliasing (MSAA) by computing an anti aliased pixel from several texel / geometry samples. The main difference between ATI and NVIDIA is the sample pattern, the amount of modes each card supports and gamma correction or adjustment. You can see the sample pattern for each AA mode on both cards (the ones' selectable from the driver control panel) below.

ATI (from left to right - 2x, 4x, 6x)



NVIDIA (from left to right - 2x, 2xQ, 4x, 6xS)



You can see that for the most part, ATI's sample pattern for 2x and 4x are similar to NVIDIA's - a mirror of each other. Both vendor's 4x sample pattern use a rotated grid sample pattern. The only major difference between ATI and NVIDIA is ATI's 6x mode and NVIDIA's 8x mode sample pattern. NVIDIA uses two texture and 8 geometry samples while ATI only uses one texture and 6 geometry samples. ATI's sample pattern looks less 'ordered' than NVIDIA so this may produce better image quality. Of course, we have to see actual in game screenshots to see whether or not that's the case. There's at least two (or three modes) we can compare directly - ATI's 2x to NVIDIA's 2x and 2xQ (short for Quincunx) and 4x implementation from both vendors. From the looks of it, 4x AA screenshots should be fairly close, if not identical. As is often is the case, more samples are more effective so expect 4x AA to be 'better' at removing aliasing than 2x. We'll mainly concentrate on 4x AA for our comparison (since this mode is the one most heavily used) and touch 2x, 2xQ, 6x and 8xS along the way.

Although comparing anti aliasing implementation is fairly easy, comparing anisotropic filtering between ATI and NVIDIA (or on this case, the Radeon X800XL and GeForce 6800) prove to be more difficult. Unlike anti aliasing, there's a lot more optimization involved. Both vendors have implemented optimizations differently so getting the same exact image from both cards is close to impossible. Both vendors implementation supports up to 16 samples anisotropic filtering and of course, plain bilinear and trilinear filtering. The MIP level patterns from the X800XL and 6800GT is very similar, although you can see NVIDIA is slightly more aggressive with their MIP levels. You can see their angle optimizations more clearly (in 8x and 16x shots).

ATI's default quality (anisotropic filtering is application controlled) (from left to right - 1x, 2x, 4x, 8x and on the bottom 16x)



NVIDIA high quality (anisotropic filtering is application controlled) (from left to right - 1x, 2x, 4x, 8x and on the bottom 16x)



By default, the image quality setting is set to 'Quality' (and not High Quality) in NVIDIA's drivers. In this settings, some optimizations are enabled (for better performance). You could still control the sample number for anisotropic filtering from the application (if it has an anisotropic filtering setting). In the MIP levels shots, you can see this setting is even more aggressive than the High Quality settings.

NVIDIA quality (anisotropic filtering is application controlled) (from left to right - 1x, 2x, 4x, 8x and on the bottom 16x)



If you were to select the number of samples for anisotropic filtering in the driver panel, the MIP levels stay fairly the same. However, there is one significant difference. Forcing the number of samples from the driver panel in Quality mode actually forces bilinear filtering on all texture stages except stage 0. You can see the difference below, the left is 16x AF High Quality and the right is 16x AF Quality.



Why is this happening? Anisotropic optimizations. Under 'Quality' image settings, whenever you change the amount of samples to filter to another option than 'Application Controlled', NVIDIA's drivers will enable all optimizations - trilinear, anisotropic sample and mip filter optimizations. In their words,

"Aniso mip filter optimization enables the NVIDIA display driver to substitute point-mipmap filtering for linear-mipmap filtering on all but the primary texture stage...Anisotropic sample optimization enables a variety of sample-relate optimizations on all but the primary texture stage...Trilinear optimization...enables the NVIDIA display driver to apply trilinear optimization for better texture filtering performance."

You can manually turn off anisotropic filter optimization to re-enable 'optimized' trilinear filtering on all texture stages for best image quality, but if you want to avoid texture shimmering, you must turn off all these optimizations. NVIDIA does have newer drivers implementing these optimizations without causing texture shimmering (the recently released 81.85), but we've already finished testing when that driver came out, so we're sticking to Forceware 78.01 for this comparison.





How does ATI compare in this regard to NVIDIA? Let's look at ATI's MIP levels.



Looks exactly like their default settings (application controlled anisotropic filtering). However, when we look at the texture stages other than 0, we see that ATI is effectively using bilinear filtering on but the primary stage (stage 0).



So technically, the only way you can get close to 'apples to apples' comparison between ATI and NVIDIA is to enable High Quality settings for NVIDIA and leave anisotropic filtering option to the default 'Application Preference' in ATI. The number of samples should be chosen by the application itself (usually through an anisotropic filtering option). Unfortunately, not all games supports this feature. Forcing anisotropic filtering means ATI has a (theoretical) upper hand since their card is doing mostly bilinear filtering. However, before drawing any conclusions, there are some points you need to remember: First, we have not seen any in game screenshots (yet) that displays lesser image quality or artifacts such as texture shimmering on ATI hardware. Two, not all games use more than one texture stage (though some do). Let's move on to actual in game screenshots.

The Real World

Let's move on the the in game screenshots, which you can find below. The left image is without AA and AF, while the right is with 4x AA and 16x AF. If you want to see the full 1024 x 768, 32 bit screenshots, just left click on the screenshots. We've also included the GeForce 6800GT AA and AF shots under 'Quality' settings in addition to "High Quality'.

Call of Duty

Radeon X800XL



GeForce 6800GT



Of course, the more interesting comparison is the AA and AF shots. We could see NVIDIA's High Quality offers the better image quality here. Just look at the texture artifacts on the right side plank (near the fence). ATI's forced anisotropic filtering at 16xAF still has some artifacts, while NVIDIA's Quality settings has the most artifacts of the three. The second difference is the fence on the upper left corner. Of course, these differences are actually minor, but they do correspond well to what we know from the mip levels diagnostics. Keep in mind though, these differences are not really noticeable in gameplay.

F1 Career Challenge

Radeon X800XL



GeForce 6800GT



The differences are very subtle, so they're not very noticeable in gameplay. Take a close look at the middle left (near and on the tarmac). You'll see the Radeon X800XL image has more blur compared to GeForce 6800GT's image, both under Quality and High Quality. The same blur can be seen on the ground and the tarmac on the left.

Nascar 2003

Radeon X800XL



GeForce 6800GT



A quick glance will not be sufficient here, since both cards produces image quality very close to each other. Looking closely at the text logo on the two cars, you'll that the X800XL does a better job under AA and AF. You could see the fence is also sharper on the X800XL screenshots compared to GeForce 6800GT's Quality and High Quality. To be honest, the differences here is minimal and certainly not noticeable in gameplay.

Homeworld 2

Radeon X800XL



GeForce 6800GT



You have to look real close (and we do mean REAL close), to see the difference between the three AA and AF shots. Look at the upper right corner (the plateau with the shadow). You could definitely see the NVIDIA's High Quality has a sharper image (without the artifacts seen on the X800XL), but the Quality settings is more blurred. On the upper left of these shots (still on the plateau), you could see High Quality does a slightly better job. Image quality under NVIDIA's Quality settings is similar to the X800XL here. However, remember that Homeworld 2 is an RTS game, where much of the gameplay is not played in this mode. Most of the time, you'll be looking at these from a distance, so these differences are not really noticeable in gameplay. We can see 4x AA between the X800XL and GeForce 6800GT is pretty much the same - or are they?

Aliasing in motion
Radeon X800XL
no AA
2xAA
4xAA
6xAA

GeForce 6800GT
no AA
2xAA
2xQ AA
4xAA
8xSAA

After looking at these image sequences, you can surely see that ATI's 4x is more effective than NVIDIA's. Even throwing more samples (8xS) is useless. Oddly enough, it is NVIDIA's 2xQ mode that's up to the task. So, NVIDIA's 4x is definitely not the same with ATI's 4x. If we were to rely on static screenshots alone, we would not be aware of this.

Star Wars: Knights of The Old Republic

Radeon X800XL



GeForce 6800GT



Let's look at anisotropic filtering between the three shots. See the far grate on the floor on the left - you could see artifacts on NVIDIA's Quality images. The High Quality settings and ATI's 'Application Preference' has no artifacts. You can also see this on the floor grate on the right - definitely banding artifacts in Quality. This is expected since NVIDIA's High Quality and ATI's Application Preference is pretty much the same or at least have very close MIP levels. 4x AA of the two cards are pretty much the same, but let's take a look at them in motion.

Aliasing in motion
Radeon X800XL
no AA
2xAA
4xAA
6xAA

GeForce 6800GT
no AA
2xAA
2xQ AA
4xAA

Well, they are pretty similar. You could definitely see both cards does a decent job in reducing artifacts (look at the white panel on the right). However, both cards are not able to remove aliasing entirely. We still see much aliasing from both cards - look at the ceiling and the table just as we're walking towards it. Even ATI's 6x AA is not up to the task. We need to use an even higher resolution if we want to 'mask' this aliasing by giving the cards more pixels to draw to. Wait a minute, you'll notice that there's no image sequence of NVIDIA's 8xS AA. Well, unfortunately the 8xS AA mode (known to the game as 8x mode) only work correctly if you disable shadows. So, we feel there's no point in testing it since image quality is not getting any better and you have to settle for less effects.

So, while MIP levels diagnostic does provide an indication of 'better' texture filtering, we should always look at in game screenshots to determine what's acceptable to our eyes (and what's not). If an application supports setting anisotropy levels, NVIDIA's High Quality settings and ATI's Application Preference is evenly matched. For applications without support of anisotropic filtering, ATI's forced anisotropic filtering is better than NVIDIA's Quality, but not up to NVIDIA's High Quality. Even so, remember that the artifacts are very small and not noticeable (especially in motion).

For anti aliasing, we could clearly see ATI has the more effective anti aliasing implementation compared to NVIDIA. While it's not 100% effective, in some cases (Homeworld 2) it's able to remove aliasing quite effectively (even under 2x) than NVIDIA's implementation. ATI's 6x mode is also more 'compatible' with games (SW:KOTOR) than NVIDIA's 8xS. Sadly, neither mode is really that much more effective than 4x in removing aliasing. So, image purist will really have to crank up the resolution to at least 1600 x 1200 with 4x AA to 'mask' aliasing. This is more effective than using 6x or 8xS AA in 1024 x 768 or even 1280 x 960 / 1024.

Performance

We've established that image quality from the NVIDIA GeForce 6800GT and ATI Radeon X800XL is only very close with application controlled anisotropic filtering. Unfortunately, not all games support this feature (both old and new). There will always be differences in image quality from both cards (and vendors) - it's up to you to decide whether or not these differences are acceptable. So, remember these facts while you're looking at the performance numbers. To better present what these cards have to offer, we include performance results for both NVIDIA's High Quality and Quality in addition to ATI's default application preference (whenever possible) or forced anisotropic filtering. There's also an fps progress graph and fps distribution table. So without further ado, here are the results.

Our test setup
AMD Athlon 64 3000+ socket 939
2 x 256 MB Kingston KVR 3-3-3 PC3200 DDR-SDRAM
MSI K8N NForce 4 SLI motherboard
Maxtor DiamondMaxPlus9 80 GBs Serial ATA 8 MB buffer
ASUS E-616 DVD-ROM
450 watts ATX power supply

Windows XP Professional with Service Pack 2 installed
ATI Catalyst 5.9 reference driver
NVIDIA Forceware 78.01 and 66.93 reference driver
NVIDIA NForce 4 6.66 reference driver
Creative SoundBlaster Live! 24 bit 5.12.1.512 driver.
DirectX 9.0c

The results:

We see a very different picture than if we were to use the built in timedemo tool. Unlike previous benchmarks scores, the Radeon X800XL is faster with AA and AF, simply running ahead of the GeForce 6800GT in this game. Of course, the image quality is somewhere between High Quality and Quality settings from the NVIDIA card, but the performance lead is significant - not bad. To be honest, any difference in image quality is not really noticeable in motion. All of these cards have enough pixel processing power to play this game with the highest detail with AA and AF enabled at the highest resolution.

	GeForce 6800GT				Radeon X800XL
	Quality		High Quality		Default
	Default	4x AA 16x AF	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0	0	0
30-45 fps	0	0	0	0	0	0
45-60 fps	0	0	0	0	0	0
60-90 fps	1	1	1	2	4	4
90-120 fps	6	8	7	9	5	6

Here we can visually see how much frame rate we're losing with both cards. The GeForce 6800GT performance's is about the same under both Quality and High Quality settings (with only trilinear optimizations is enabled). With AA and AF (both aniso mip filter and sample optimizations enabled), we see a pretty big hit under High Quality settings. On the other hand, the Radeon X800XL does not lose as much (or gain as much) when AA and AF is enabled.


Here we see a different story. With forced anisotropic filtering, the Radeon X800XL is barely reaching the performance of GeForce 6800GT running under High Quality and Quality settings. On higher resolutions, the X800XL performance is at par with the GeForce 6800GT in High Quality. However,it is interesting to note that the Radeon X800XL continues to have the faster minimum fps on all resolutions and still manages a 38 fps minimum with AF on 1600 x 1200.

	GeForce 6800GT				Radeon X800XL
	Quality		High Quality		Default
	Default	4x AA 16x AF	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0	0	0
30-45 fps	0	0	0	0	0	0
45-60 fps	0	0	0	3	0	0
60-90 fps	22	24	21	22	21	22
90-120 fps	9	7	10	6	10	9

Just like in Call of Duty, we see the GeForce 6800GT experience some performance penalty under High Quality compared to Quality. The difference grew larger when AF is enabled (through the drivers). Compared to the GeForce 6800GT, the Radeon X800XL lost very little performance - even if you compared it with NVIDIA's Quality settings.


In this system limited benchmark, we're seeing the Radeon X800XL is faster by at least 10% overall. It's not until we hit 1600 x 1200 that the benchmark becomes graphics limited (with AA and AF enabled). In 1600 x 1200, we see the GeForce 6800GT under Quality outpaced the Radeon X800XL . So, while overall the Radeon X800XL is the faster card here, it's slightly bested by the GeForce 6800GT in 1600 x 1200 with AA and AF (or so it seems).

	GeForce 6800GT				Radeon X800XL
	Quality		High Quality		Default
	Default	4x AA 16x AF	Default	4x AA 16x AF	Default	4x AA 16x AF
<30 fps	0	0	0	0	0	0
30-45 fps	0	0	0	16	0	0
45-60 fps	19	23	19	45	0	25
60-90 fps	59	95	59	57	45	93
90-120 fps	41	0	40	0	73	0

We use 1600 x 1200 for the two graphs, the game being graphics limited on 1600 x 1200 or higher. Overall, we see the two cards acting very similar - with AA and AF, there's quite a significant performance hit. The Radeon X800XL has a slight edge, since it's more efficient with AA and AF compared to the GeForce 6800GT. It spent less time in the 45-60 fps range than the GeForce 6800GT under High Quality settings).


In Brothers in Arms, we see the GeForce 6800GT taking a very commanding lead over the Radeon X800XL - under both Quality and High Quality settings. The Radeon X800XL is not even reaching the minimum 30 fps mark, even without AA and AF. Now, that's odd, since with application controlled settings, it should be at least similar to the GeForce 6800GT under High Quality settings.