Tech-Hounds.com

Because gamers play games, not benchmarks




A Look at the GeForce 8800 GTS

PCI Express x16 graphics cards
Chipset GeForce 8800GTS 640, 320 MB, 384 bit

The launch of any high end graphics card certainly attracts lots of attention from users particularly gamers, reviewers not to mention IHVs and ISVs like game developers and publishers. There is only one thing that's bigger - the launch of a graphics card based on a new API. The GeForce 8800 certainly fills that spot. Users, reviewers and quite likely, competitors were caught off guard when NVIDIA launch their latest generation of card, the GeForce 8800 series, codenamed G80.

The biggest surprise is off course unified shaders support, one of the major 'attraction' or technological advances, if you prefer, with Microsoft's newest incarnation of DirectX, DirectX 10. For years, NVIDIA have argued against using unified shaders in their design, claiming that there is no immediate need for such a move. Yet, here and now, the GeForce 8800 series made up of three offerings - the GeForce 8800GTS, GeForce 8800GTS 640 MB and GeForce 8800GTS 320 MB based on the same design all utilizes a unified shader architecture instead of the traditional vertex / pixel shader units arrangement.

The GeForce 8800 series is certainly poised to further NVIDIA's lead over competitor ATI, now the graphics division of AMD. Determined not to make the same mistake as they did with the transition from DirectX 8 to DirectX 9, NVIDIA chose to employ the tried and true 90nm fabrication process for the GeForce 8800, instead of the more risky 60nm process. Sticking to a tried and true process means NVIDIA will be able to launch their products much sooner than ATI. Other lessons learned from the last transition are also taken to heart. NVIDIA chose to follow DirectX 10 specification as close as possible, working closely with Microsoft to tweak and tune G80 design so there will be no 'growing pains'  and performance issues later on.

Much like ATI's Radeon 9700 Pro, the GeForce 8800 series is ahead of its time - meaning though its already available in the market for some time, software support was practically non existent. Part to blame is Microsoft and their decision to not just employ a new driver model for the next version of Windows, Windows Vista, but also exclusively bundle DirectX 10 only with Vista. That means users with previous versions of Windows will not be able to fully tap the hardware capabilities of any DirectX 10 compliant graphics card such as the GeForce 8800. Even now, with Vista available, there's practically no DirectX 10 game titles available on the market, at least until the end of the year.

Vista should have become an important priority, one that NVIDIA seem to overlook. Due to time and resource constraints, NVIDIA chose to focus driver development to assuring GeForce 8800 compatibility and performance with the currently available version of Windows, Windows XP. They also chose to focus on SLI and providing SLI profiles for current and soon to be release game titles. So, Vista driver development took a back seat, a decision that will cause much grief for early Vista users.

It turned out the new Longhorn Device Driver Model (LDDM) caught many manufacturers off guard, but the problem seem far worse with NVIDIA hardware, and not just with the GeForce 8800. Volatile performance, compatibility issues and even Blue Screens of Deaths greeted users of NVIDIA's GeForce 7 and 8 series upgrading their Windows XP installations to Vista. Though recent driver development shows promise - performance beginning to be where it should be, no more crashes and the like - there's still relatively few PC vendor who will bundle Vista with their NVIDIA hardware equipped systems.

DirectX 10 and Vista

Undoubtedly, the most hyped reason to buy the GeForce 8800 is its support for DirectX 10 and Windows Vista. As many early bird users have found, life with a GeForce 8800 series of card in Windows Vista can be 'a major pain'. Thankfully, the newer drivers seems to improve the situation quite a bit.

As for DirectX 10 itself, since there's actually no DirectX 10 title on the market, we think it's a moot point to buy a card based on what 'it can do'. It's more justifiable to buy a card for what 'it can do' for current titles. Thankfully, users should be able to see DirectX 10 patches for some DirectX 9 titles to show up in the coming months. These will have to do until we see a game that's really built to take advantage of what DirectX 10 can offer, and that's is still a long ways off.

You may be asking why do we say that. It's important to understand that Vista does not only bring DirectX 10 to the table, but also a new driver model. So, a game or application may still use DirectX 9 shaders, be it SM 2.0 or SM 3.0, however developers may be able to use DirectX 10 to 'alleviate' many technical hurdles and problems that is present in DirectX 9. Many developers and publishers have promised to update their previous titles with DirectX 10 support. It's very likely these applications are not 'built from scratch' for DirectX 10. In that line of thinking, these applications are less likely to push DirectX 10 features and capabilities of a card like the GeForce 8800 series. So, we prefer to wait for a 'real' DirectX 10 game title or benchmarks and use what games are available today for this article.

Image Quality Comparison

Before looking at the performance the GeForce 8800 GTS has to offer for current titles, first we'd like to take a closer look at image quality. In the past, NVIDIA have come under fire for numerous image quality problems and compromises -  'brilinear' filtering and shimmering textures. To be fair, the aforementioned problems can be avoided or at least controlled to a minimal degree by always using 'High Quality' for Image Quality settings in the Forceware drivers control panel.

By default, the Forceware drivers still sets Image Quality settings to 'Quality' for the GeForce 8 series. With 'Quality', Trilinear optimization is enabled. So, naturally we chose to set Image Quality settings to High Quality to disable Trilinear and Anisotropic sample optimizations. This should offer the best image quality settings for these cards. For comparison sake, we used a GeForce 7950GT (also with "High Quality' image settings) and a Radeon X1950 Pro with both default and 'High Quality' anisotropic filtering enabled.

Anisotropic Filtering

Reviewers and general graphics geeks should be familiar with the screenshots below. The shots are taken from D3D AF Tester, showing MIP map boundaries with different texture filtering options. We change the default values a little bit, opting to use much lower tunnel faces to angles you'll most likely see inside a game. The shots are arranged consecutively in this order - 1x AF, 2x AF, 4x AF, 8x AF, and 16x AF.

GeForce 7950GT

       

Radeon X1950 Pro

       

Radeon X1950 Pro (High Quality AF)

       

GeForce 8800GTS

       

At 1x and 2x AF, the Radeon X1950 Pro has the more conservative MIP boundaries, while the GeForce 7950GT and 8800 are practically the same. At 4x AF, the three started to differ - the Radeon X1950 Pro is more aggressive here, the GeForce 8800 is most conservative while the GeForce 7950GT is somewhere in between. We can already see the tell tale signs of angle optimizations on both the Radeon X1950 Pro and GeForce 7950GT but not on the GeForce 8800. At 8xAF, the GeForce 8800 definitely offers the best in regards to image quality filtering wise. The Radeon X1950 Pro and GeForce 7950GT differs slightly. On the upper parts of the tunnel, the GeForce 7950GT is more conservative. 16x AF shots of these two cards are not that different from 8x AF.

Now let's focus on the Radeon X1950 Pro HQ AF and GeForce 8800 AF shots. At 1x, 2x and 4x AF, the Radeon X1950 Pro still uses more conservative MIP boundaries, but we can also see there's already signs of some angle optimizations in effect. Not much, but it's there. At 8x and 16x AF, the Radeon X1950 Pro HQ AF basically 'switched' to using more aggressive MIP boundaries for the upper parts of the tunnel.

Judging by these shots, the GeForce 8800 should offer much higher image quality - filtering wise - when compared to the Radeon X1950 Pro default AF and GeForce 7950GT. The Radeon X1950 Pro put up a good fight with HQ AF enabled, the difference should be very hard to notice in real word situations. Or is it?

Angle Independence

Let's test that assumption. We set D3D AF Tester to 16x AF, but now we raise the number of faces on the tunnel to 9. An odd number you might say, but at this setting allows us to 'test' 9 different angles. Let's see how the cards fare in this test.

GeForce 7950GT



Radeon X1950 Pro



Radeon X1950 Pro (High Quality AF)



GeForce 8800GTS



The Radeon X1950 Pro is definitely more aggressive than the GeForce 7950GT at default settings. But of course, it's the Radeon X1950 Pro HQ AF and the GeForce 8800 AF shots we are more interested in. Again, the GeForce 8800 shows little variations between MIP boundaries on different angles, unlike in the Radeon X1950 Pro HQ AF shot.

Quality vs High Quality

GeForce 8800GTS

       

For the curious, we also did some test with the default 'Quality' image settings on the GeForce 8800. As you can see for yourself, angle independent filtering is still enabled in this mode. Though trilinear optimization is enabled, you should be able enjoy better image quality overall in this setting with the GeForce 8800 than on GeForce 7950GT and Radeon X1950 Pro.

Anti Aliasing

The DX9 FSAA Viewer allows us to see not only the number of samples used when anti aliasing is enabled, but also the pattern they are organized. Modern cards uses some modified form of Rotated Grid multisample anti aliasing (MSAA), which offers much better results quality wise than Ordered Grid MSAA.

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

   

GeForce 8800GTS (CSAA)

 

GeForce 8800GTS 16xQ



With the GeForce 8800 and quite possibly the whole GeForce 8 series, NVIDIA introduce a new form of multisample anti aliasing to the masses - Coverage Sampling Anti Aliasing or CSAA for short. Like the traditional form of MSAA, CSAA still uses the same number of color samples, but adds more 'coverage samples'. We'll get more into this in a while.

The interesting comparison in this test is how the GeForce 8800 uses different sample patterns for its 'true' 8x sample MSAA than NVIDIA's previous generation - the 8xS anti aliasing mode. With a much less 'ordered' sample pattern, the 8xQ MSAA seem to offer an improvement in image quality, but the use of 2 texture samples in super sampling in 8xS may still offer the better image. The trouble with 8xS is that since it uses supersampling, performance is usually poor with current game titles. Because it relies on multisampling, 8xQ MSAA should offer less of a penalty hit than 8xS but we'll have to see if the image quality is worth switching from 4x MSAA. 

Coverage Sampling Anti Aliasing in Brief

What is Coverage Sampling Anti Aliasing? Well, lets look at NVIDIA's description of CSAA (taken from NVIDIA's CSAA PDF brief, here):

"In summary, CSAA produces antialiased images that rival the quality of 8x or 16x MSAA, while introducing only a minimal performance hit over standard (typically 4x) MSAA. It works by introducing the concept of a new sample type: a sample that represents coverage. This differs from previous AA techniques where coverage was always inherently tied to another sample type. In supersampling for example, each sample represents shaded color, stored color/z/stencil, and coverage, which essentially amounts to rendering to an oversized buffer and downfiltering. MSAA reduces the shader overhead of this operation by decoupling shaded samples from stored color and coverage; this allows applications using antialiasing to operate with fewer shaded samples while maintaining the same quality color/z/stencil and coverage sampling. CSAA further optimizes this process by decoupling coverage from color/z/stencil, thus reducing bandwidth and storage costs."

On the GeForce 8800, the following CSAA modes are supported:

CSAA Mode Number of Color/Z/Stencil
Samples
Number of Coverage Samples
8x 4 8
8xQ (Quality) 8 8
16x 4 16
16xQ (Quality) 8 16

So, 8xQ CSAA is the only 'true' 8x MSAA (in the traditional sense), using 8 color / Z / stencil samples, unlike 8x which uses 4 color / Z / stencil samples but 8 coverage samples. We have a little gripe with NVIDIA - we think it would be more 'correct' to simply use 8x for 8xQ instead and different ones for the other CSAA modes (for example 4x8 for 8x CSAA, 4x16 for 16x CSAA and 8x8 for 16xQ). Such an arrangement will likely cause less confusion for average users, who will likely assume 4x (MSAA) and 8x (CSAA) employ the same method. In our opinion, calling NVIDIA's GeForce 8800 8x or 16x CSAA as 8 or 16 samples anti aliasing, respectively,  is 'bending' the truth quite a bit, some might even say its misleading. Let's see if coverage sampling anti aliasing (CSAA) offer something 'more' compared to traditional multi sample anti aliasing.

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

     

This small application from ToMMTi-Systems shows a unique perspective on anti aliasing. With it, we can see how effective anti aliasing implementation on lines with different slopes. If only there was also support curves in addition to lines, it would just be perfect. Anyway, back to the shots. One of the easiest spot to see differences is the third quadrant of the shot - between 6 and 9 o clock (180 to 270 degrees). At 4x MSAA, the Radeon X1950 Pro and GeForce 8800 was able to display the entire array of lines, while the GeForce 7950GT seems to 'lost' two lines (or triangles if you prefer). If you look REAL close (or zoom in), you'll see the Radeon X1950 Pro shot displays less aliasing in this area, even compared to the GeForce 8800. Moving to the center, you'll also see a more 'complete' picture on both the Radeon X1950 Pro and GeForce 8800 - no loss of details here.

Now, let's focus on 8xS, 8x CSAA an 8xQ MSAA shots. The GeForce 7950GT with 8xS still 'lost' some detail compared to 8x CSAA and 8xQ MSAA on the GeForce 8800. Good or bad, well that depends. Look at the whole picture and you'll definitely see 8xS does a better job in reducing aliasing overall. Both approaches have disadvantages, you either lost details (8xS) or have more aliasing (8xQ and 8x CSAA). 8x CSAA and 8xQ MSAA shots are actually pretty close to each other, though 8xQ seem to offer a more 'distinctive' image - you'll notice more pixels near the center with 8xQ MSAA and 8x CSAA.

Real World Implementation

We've seen the GeForce 8800 series has to offer in terms of anisotropic filtering and anti aliasing, but how does it translate to the real word? Let's see how they do in games. We admit, we got a 'little' carried away with these game screenshots (there's a whole lot of them) so if you want to skip them, click here.

Still with us? Thank you, now let's take a look at some shots.

Call of Duty

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

         

If you look at the 4x AA and 16x AF shots between the Radeon X1950 Pro and the GeForce 7950 and 8800, you'll notice that the wood on the plank / platform on the right is sharper on the GeForce cards. The same is true for the wood plank on the far lower platform. The far left part of th fence also seem to be 'missing' on the Radeon X1950 Pro shot. You'll also see less aliasing on the scope and bolt of the rifle with the GeForce 7950GT and 8800.

With 8x anti aliasing, there is one glaring difference between the shots between the 8xS on the GeForce 7950GT and 8x CSAA, 8xQ, 16x CSAA and 16xQ on the GeForce 8800. The far fence - wherever transparent textures are used. It would seem the biggest winner, anti aliasing wise, in this game would be 8xS on the GeForce 7950GT.

Homeworld 2

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

         

Although the shots don't quite line up with each other, they're still within the same ball park. The easiest spot to see aliasing between 4x AA 16x AF shots is on the center of the image - the 'nudge' on the upper part of the station (above the interceptor on the GeForce 8800 shot). With motion, the aliasing is very noticeable on the Radeon X1950 Pro, but less noticeable (though it can still be seen clearly on the GeForce 7950GT and GeForce 8800).

With 8x anti aliasing, we see very little difference between 8xS, 8x CSAA, 8xQ, 16x CSAA and 16xQ. 4x AA seem to be enough in this scene. The GeForce cards does a much better job minimizing aliasing in this shot than the Radeon X1950 Pro. However, in our aliasing in motion test, the GeForce 7950GT falls flat against the Radeon X1950 Pro and GeForce 8800. You can see it if you look at the slideshow below.

Aliasing in motion (.bmp files in RAR package)

GeForce 7950
default
4x AA
8xS AA

Radeon X1950 Pro
default
4x AA

GeForce 8800GTS
default
4x AA
8x CSAA
8xQ AA
16x CSAA
16xQ AA

8xS on the GeForce 7950GT does a lot better job than the card's 4x MSAA, but there's still some aliasing near the end. Its just about as effective as traditional 4x MSAA on the GeForce 8800. With 8x CSAA and 8xQ, the GeForce 8800 was able to take it up a notch, reducing that little bit of aliasing near the end, with 16xQ offering the best results - at par with 4x MSAA on the Radeon X1950 Pro. So, it would seem the card that's more effective in dealing with aliasing here is the GeForce 8800. It may still exhibit some aliasing at 4x MSAA, but it does well on both scenes while the Radeon X1950 Pro exhibits more noticeable aliasing on the on the first test.

Nascar 2003

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

         

Let's look at 4x AA and 16xAF shots first. Both GeForce cards were able to preserve more detail in the background. Impostor (background / skybox) textures are much sharper than on the Radeon X1950 Pro. The Radeon X1950 Pro lost some ground here, but it makes up for it by showing less aliasing on the furthest part of the track. Look at the race car right in front, you'll see the word 'Goodwrench.com' is much sharper on the Radeon X1950 Pro than on both GeForce cards. However, if you look at the car on the left, the lettering on the door is blurry on the Radeon X1950 Pro and sharper on both GeForce cards. The buckle for the side nets are also much sharper than on the Radeon X1950 Pro. The only noticeable difference between the GeForce 7950 and 8800 seem to be the crowd on the grandstand, most likely due to angle independent anisotropic filtering.

Now, let's compare shots with 8 samples anti aliasing (8xS, 8xQ and 8x CSAA). It's worth noting 16x CSAA seem to exhibit the same problem like the Radeon X1950 Pro with the car and the buckle on the left. Moving on, on the furthest part of the track, you'll see less aliasing with 8xS and 16xQ, but the writing "Goodwrench.com" is more discernible with 8xS on the GeForce 7950GT. So, overall, we think the GeForce 7950GT is has the more effective anti aliasing with 8xS. It has sharper background textures, less aliasing overall although it loses some sharpness and details on the crowd texture to the GeForce 8800.

SW: Knights of the Old Republic

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

     

At 4x AA, the GeForce 7950GT and 8800 looks pretty much the same. Understandably, there are visible differences when the two are compared to the Radeon X1950 Pro shot. Both GeForce cards seems to be better with dealing aliasing on faraway objects (the table in the next room and the far doorway arch) but the Radeon X1950 Pro does a better job with closer objects (white panel on the right, next to the door and the seats right in front of the player). However, on the grate on the left wall, the GeForce 7950GT has the better anti aliased image.

KOTOR is unique in that it is a game that natively recognizes 8x anti aliasing if the hardware supports it. 8x anti aliasing corresponds to 8xS on the GeForce 7950GT and 8xQ on the GeForce 8800. So, let's try to focus on the differences between these modes and 8x CSAA. First, the grate on the left - 8xS is best, 8xQ is all right and 8x CSAA is the worst. Another area to spot is the left side of the door arch. We think 8x CSAA works best here, because it was able to keep sharpness while removing aliasing. 8xS is rather blurry and there's still some aliasing with 8xQ. The seats also look better (less aliasing) with 8x CSAA. For faraway objects, it would seem 8xS work best.

Aliasing in motion (.avi captures - using Xvid)

GeForce 7950
default
4x AA
8xS AA

Radeon X1950 Pro
default
4x AA

GeForce 8800GTS
default
4x AA
8xQ AA

We apologize for the quality of the GeForce 7950GT 4xAA recording. There's seem to be a problem with the latest Forceware drivers and Hyperionics Hypercam on the GeForce 7950GT. Rest assured, no such display corruption appear when you're playing the game. If you look at the three videos, the only aliasing in motion you'll see is very likely on the top part of the far corridor and the door arch and the table in the next room. Neither card 4x MSAA did any good here, later on we'll see with a higher resolution such as 1280 x 1024.

Looks like there's still a problem with using 8x anti aliasing, be it 8xQ or 8xS with SW: KOTOR. You definitely see aliasing on the white panel, next to the door. To be fair, we did notice less aliasing on the table with 8xQ and on both the table and the far corridor / door arch with 8xS. It's a moot point though, since using 8x anti aliasing seem to introduce artifacts in other areas.

Serious Sam II

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

         

4x MSAA between the three cards are about the same. There's seem to be very little difference between 8xS, 8xQ, 8x CSAA and 16xQ in this shot.

GTR

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

         

The immediately noticeable difference between the three cards with 4x MSAA is the grandstand. The Radeon X1950 Pro and the GeForce 7950GT was unable to filter the entire grandstand, something that is not an issue with the GeForce 8800. Angle independent filtering at play? Seems likely. Moving on to aliasing, we think the Radeon X1950 Pro has the upper hand here - though by a very slight margin. Although the GeForce 7950GT and 8800 has a 'smoother feel' - less aliasing - in their shots, it loses too much 'sharp' details. Take a good look at the writing "John Guest" on the black Porsche directly in front the car. It's definitely is sharper on the Radeon X1950 Pro shots than on both GeForce's shots.

Using 8xS seem to 'cure' the lack of detail on the grandstand on the GeForce 7950GT. There's also less noticeable aliasing on the fender, the hood, wiper and spoilers of the cars. Again, we found ourselves leaning to the GeForce 7950GT's 8xS as the better anti aliasing implementation compared to 8x CSAA and 8xQ MSAA on the GeForce 8800.

Need for Speed: Carbon

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

         

At a glance, there doesn't seem to be any difference between the three cards. However, look close enough and you'll soon find bits here and there. The easiest to spot is the road markings just in front of the car. See how the lines became less clear / disconnected on the GeForce 7950 and 8800 as it got further and further? The Radeon X1950 Pro definitely does the better job here. Now focus on the car itself - this time notice how the highlights on the GeForce 8800 has less noise - aliasing artifacts - compared to the GeForce 7950 and Radeon X1950 Pro shots. You can see it on the side and the spoiler. However, on the bottom part of the fender. we think the GeForce 7950 does a better job.

While the 8xS shots may look nice, let's be honest - there's no way you're going to use 8xS in NFS: Carbon, because of performance reasons. You might try it with a GeForce 7950GT SLI setup or a 7950GX2, though we doubt its worth it. The road marks looks exactly the same with the GeForce 8800 shots in various AA modes - the only difference we can see is on the car. The highlights definitely looks better with 8xQ - no disconnected highlights on the spoilers. Both 8xQ and 16xQ don't have 'noisy pixels' on the left side of the car (near the window). The GeForce 7950GT 8xS shots is actually not only pretty close to the Radeon X1950 Pro shot, but also feature less aliasing on the highlights like the GeForce 8800 8xQ shot. Too bad frame rate was way too low to be of any use - 15 to 20 fps.

Aliasing in motion (.avi captures - using Xvid)

Radeon X1950 Pro
default
4x AA

GeForce 8800GTS
default
4x AA
8xQ AA

Because of performance reasons, we couldn't capture the game with AA enabled on the GeForce 7950GT. So again, this time we're going to focus on the Radeon X1950 Pro and the GeForce 8800G. The highlights are probably the easiest spot to see aliasing. Both cards are at par with each other here in the first part of the video  - none can really eliminate all aliasing though they are noticeably better than no anti aliasing at all. It's not actually until we're viewing the same angle on the screenshots that we begin to see a difference. The GeForce 8800 does better with the highlights on the spoilers with 4x and 8xQ than the 4x on the Radeon X1950 Pro. Now that you know where, to look, you'll see the difference become more pronounced as the car lurches forward.

The Elder Scrolls IV: Oblivion

GeForce 7950GT

   

Radeon X1950 Pro

 

GeForce 8800GTS

   

There's actually very little difference separating the three cards in this game with 4x MSAA. We also don''t see that much difference between 4x, 8xQ and 8x CSAA on the GeForce 8800.

Let's recap what we've seen so far. Overall we think 4x AA quality on the GeForce 8800 is better than 4x AA on the GeForce 7950GT. A much needed improvement, we might add. NVIDIA have stepped up to the plate and offer the better image quality - at least compared to previous generation of cards from both manufacturers. We saw the evidence on both the ToMMTi-Systems test app as well on games - our chosen scenes from Homeworld 2 comes to mind. In regards to anisotropic filtering, we saw noticeable improvements In both Nascar 2003 and GTR2 on the GeForce 8800, very likely due to angle independent anisotropic filtering.

However, we didn't see any noticeable improvement with 8xQ or 8x CSAA and their higher samples counterparts over the more traditional 4x MSAA . Well, maybe except in Homeworld 2 and Need for Speed: Carbon. We actually found ourselves wishing NVIDIA had retained 8xS for the GeForce 8800 - we see definite improvements with 8xS compared to 8xQ and 8xCSAA (even 16xQ and 16xCSAA) in Call of Duty (transparent textures), Nascar 2003 and GTR2. The road marks on NFS: Carbon, slightly more 'blurry' textures in Nascar 2003 and GTR2.

Real World Implementation - Higher Resolution (1280 x 1024 and 1280 x 960)

What if we raise the resolution? With higher resolution, more pixels are drawn per screen area which should help multisampling when compared to the hybrid multisampling / supersampling of 8xS. Alright, let's see what we can gleam with slightly higher resolution. We chose 1280 x 960 for Homeworld 2 and 1280 x 1024 for Nascar 2003, GTR2 and Need for Speed: Carbon. We wanted to see if raising the resolution will help the alleviate some of the aliasing still visible on the many anti aliasing modes on the GeForce 8800.

Homeworld 2

GeForce 7950GT

 

Radeon X1950 Pro



GeForce 8800GTS

   

In this scene we actually didn't see that much of a difference at 1024 x 768 between 4x MSAA on both the GeForce 7950GT and 8800 and the many extra anti aliasing modes on the GeForce 8800. We still didn't see any difference between these modes, but the good news is that we did see less aliasing on all of  these modes at 1280 x 960. The same also holds true for the Radeon X1950 Pro. There's still aliasing but its far less noticeable now.

Aliasing in motion (.bmp files in RAR package)

GeForce 7950GT
4x AA
8xS AA

Radeon X1950 Pro
4x AA

GeForce 8800GTS
4x AA
8x CSAA
8xQ AA

With higher resolution (1280 x 960), the GeForce 8800 with just 4x MSAA was able to match the quality of the Radeon X1950 Pro 4x MSAA. Naturally, 8xQ and 8x CSAA offers the same quality - no aliasing are visible. The GeForce 7950GT with 8xS was still not able to fully eliminate aliasing in this scene. So it looks like you need to use a resolution higher than 1024 x 768 to really get the benefit of anti aliasing on the GeForce 8800. Let's see if that is also true for the other games..Another thing to note is that we're still not seeing any noticeable difference between 4x, 8xQ and 8x CSAA.

Nascar 2003

GeForce 7950GT



Radeon X1950 Pro



GeForce 8800GTS

   

Well, the good news is that the "Goodwrench.com" is more discernible on the GeForce 8800 on all modes. The Radeon X1950 Pro retains the blurriness in background texture and left car / buckle problem. Obviously, the crowd on the grandstand is 'sharper' on all GeForce 8800 shots. With higher resolution, the furthest part of the track is smooth with 4x MSAA and 8x CSAA on the GeForce 8800, though curiously not with 8xQ on the GeForce 7950GT. However, it we had to nitpick, we were a little bit disappointed with the flag poles - they actually look better on the Radeon X1950 Pro 4x MSAA and GeForce 7950 8xS.

SW: Knights of the Old Republic

GeForce 7950

 

Radeon X1950 Pro



GeForce 8800GTS

   

There seem to be no difference at all between 4x MSAA on the GeForce 7950GT and 8800.

If you look at the same problems areas, you'll see 8xS does better with faraway objects and the grate on the left. It is a bit blurry compared to 8xQ and 8x CSAA. At 1280 x 1024, 8xQ seem to be able to match 8xS quality here, but not quite so with the far door arch, corridor plus the table in the next room. The quality is somewhere midway between 8xS and 8x CSAA.

Aliasing in motion (.avi captures - using Xvid)

GeForce 7950
4x AA
8xS AA

Radeon X1950 Pro
4x AA

GeForce 8800GTS
4x AA
8x CSAA
8xQ AA

The same glitch seen on the GeForce 7950GT is also present on the GeForce 8800 at 1280 x 1024. Though its hard to see because of the display corruption, the Radeon X1950 Pro seems to be better at removing aliasing than both GeForce cards. We still see some aliasing on the table with 4x MSAA on the GeForce 8800 and 7950GT.

Thankfully, no such glitches exist with the 8x anti aliasing video. Looking back and forth between these three modes, we think 8xS remains the superior mode compared to 8xQ and 8x CSAA. 8x CSAA actually looks worse than 4x, the aliasing clearly visible on the table and on the door arch. 8xQ fares much better, but there's still some aliasing left on these problem areas. With 8xS, aliasing on the table and door arch is minimal, however we saw  seeing aliasing in parts where there is none in 8xQ and 8xCSAA. Overall, we think it's actually wiser to choose 4x MSAA for this game, because you actually see less areas with aliasing.

GTR2

GeForce 7950GT



Radeon X1950 Pro



GeForce 8800GTS

   

Good news for the GeForce 8800. By raising the resolution to 1280 x 1024, we are finally seeing 4x MSAA, 8x CSAA and 8xQ shots match and surpass those of Radeon X1950 Pro 4x MSAA and GeForce 7950GT 8xS. There no visible blurriness anymore on the "John Guest" writing on the black Porsche at this resolution compared to the Radeon X1950 Pro.  There's also no more aliasing on the fender, spoilers, hood or wiper that was visible at 1024 x 768 when compared to the GeForce 7950GT's 8xS. However, we fail to see any noticeable difference between 4x MSAA, 8xQ and 8x CSAA.

Need for Speed: Carbon

GeForce 7950GT

 

Radeon X1950 Pro



GeForce 8800GTS

   

Unfortunately, we're still seeing some artifacts with the road marks on the GeForce 8800 various anti aliasing modes. It's much less noticeable now, but it is still there. The Radeon X1950 Pro still does a better job here. However, with a higher resolution we can now see a new area that's different between the shots from the three cards - the wall on the left. There's noticeably less aliasing on the the GeForce 8800 various AA modes compared to the Radeon X1950 Pro. And as before, we're seeing a much more less aliased highlights with GeForce 8xQ than 4x MSAA or 8x CSAA.

We think our conclusion stands, but now made more complete because we factor in different resolutions. 4x MSAA on the GeForce 8800 is definitely better at removing aliasing than 4x MSAA on the GeForce 7950GT. At 1280 x 1024 or 1280 x 960, the card was able to remove any visible aliasing in our Homeworld 2 scene, matching the quality previously only seen with 4x MSAA on the Radeon X1950 Pro (and surpassing it in another test scene). In GTR 2, with more pixels to play with, 4x MSAA on the GeForce 8800 can offer comparable quality to 8xS, if not the same.

There is still some areas where the GeForce 8800 4x MSAA have trouble - Nascar 2003 faraway flagpoles, car highlights on Need for Speed: Carbon for example. Upping the sample number by using 8xQ may be effective, but not always. As for 8x CSAA, well - we think quality wise, you should use 8xQ instead whenever possible. We don't think 8x CSAA offer a 'big' noticeable improvement over 4x MSAA.

The biggest gripe we have is with the GeForce 8800 various AA modes is, well, it looks like you need to use at least 1280 x 960 or 1280 x 1024 (or higher) to see 'better' image quality. Speaking of AA modes, we honestly think NVIDIA should have retained 8xS for the GeForce 8800. At the very least, it should offer another option for users who are unable (or not willing) to use resolutions any higher than 1024 x 768. 8xS also seem to be more effective with troublesome transparent textures like we saw with Call of Duty. Turning on transparency AA have no effect in this game.   

HDR plus AA and Transparency Anti Aliasing

Another notable image quality improvement NVIDIA imbued the GeForce 8800 is the support for floating point HDR effects and AA together. Remember, AMD introduced this feature a long time ago with the X1800 series - almost two generations ago (counting the release of next generation cards this year). Two games that come to mind when talking about HDR and AA are Serious Sam II and Oblivion. However, Oblivion doesn't natively supports using HDR and AA on both NVIDIA and AMD graphics cards - you have to 'force' anti aliasing through their respective drivers. So, we're relying on Serious Sam II at the moment. First off, let's see how does 4x Transparency AA with Supersampling and Adaptive AA with Quality has to offer, image quality wise in Serious Sam II.

Serious Sam II (Transparency AA)

GeForce 7950GT



Radeon X1950 Pro



GeForce 8800GTS



We think the GeForce 7950GT is more 'noisy' than the GeForce 8800 or Radeon X1950 Pro. GeForce 8800 transparency anti aliasing quality actually went 'back' to the level we originally saw with the GeForce 7800GTX and that's a good thing. Now, let's look at the HDR plus AA shots. Of course, this means we're only going to look at the Radeon X1950 Pro and GeForce 8800.

Serious Sam II (HDR plus AA)

Radeon X1950 Pro



GeForce 8800GTS



4x MSAA and 16x AF with HDR enabled looks better on the GeForce 8800 than on the Radeon X1950 Pro, at least if you're looking at the gun on the left. The gun on the right seem to favor Radeon X1950 Pro - differences due to different sample patterns perhaps? However, there's no denying textures are much sharper too with less artifacts on the GeForce 8800. Most likely the GeForce 8800 more angle independent anisotropic filtering is the cause.

While testing with HDR plus AA and transparency / adaptive AA, we noticed these cards and their respective drivers from both manufacturers will not enable transparency / adaptive AA in Oblivion if you try to force anti aliasing from the driver panel. So, users will have to trade off HDR or adaptive / transparency AA in Oblivion.

Performance

Graphical settings for these games are set to their highest possible values. Notable differences are F.E.A.R where we disable "Soft Shadows", Quake 4 where we use High Quality - not Ultra Quality - but anisotropic filtering was set to "1" for default tests and "16" for AA / AF tests. We used Serious Sam II built in "Maximum" quality preset then change resolution and AA / AF settings accordingly. All of these settings can be seen here. Our TES IV: Oblivion settings are as close as we can get to Ultra High Quality - you can see our test settings here. For Need for Speed: Carbon, we used the game's "Maximum" video quality preset, but change the filtering to "Trilinear" for default tests - AA AF tests are run with "4x" and "Anistropic". NFS: Carbon test were done with the Sprint Race on Lincoln Boulevard. You can see the settings we used for GTR 2 here. Clear weather was used for our run of Monza GP tests.

We'd like to thank Tagan for supplying the additional power supply and Gigabyte for supplying both the Gigabyte P965-DS3P and Radeon X1950 Pro for this article. And of course, thanks to ASUS for lending us their EN8800GTS 640 MB and EN7950GT also to Leadtek for the PX8800GTS-TDH 320 MB. Please keep in mind that the GeForce 7950GT and Radeon X1950 Pro are not direct competitors to the GeForce 8800GTS. Their results are displayed merely as a reference point.

Our test setup
Intel Core 2 Duo E6300 socket LGA-775 (running at 7 x 400 MHz)
4 x 512 MB A-DATA Vitesta 5-5-5-18 PC6400 DDR2-SDRAM
Gigabyte P965-DS3P Intel P965 motherboard
Maxtor DiamondMaxPlus9 80 GBs Serial ATA 8 MB buffer
LiteOn 1673S DVD-RW
Tagan TG530-U15 530 watts ATX/BTX power supply

Windows XP Professional with Service Pack 2 installed
ATI Catalyst 7.3 reference driver
NVIDIA Forceware 93.71 reference driver (for the GeForce 7950GT)
NVIDIA Forceware 158.19 BETA reference driver (for the GeForce 8800GTS 640 MB and 320 MB)
Intel Chipset Software Installation Utility 8.1.0.1006
DirectX 9.0c
all respected games used for benchmarks have been updated to their latest,final builds.

The graphs are pretty self explanatory, but in case you can't see the text, they're arranged in groups of three colors: green for minimum fps,  blue for average fps and red for maximum fps.  The first group of graphs are results taken with default settings - AA and AF is not enabled. The second group of results are results with 4x AA and 16x AF.

The results:

F.E.A.R - Performance Test, 1024 x 768
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
75
173.58
431
57 (4x AA 16x AF)
119.24 (4x AA 16x AF)
299 (4x AA 16x AF)
.
71
175.28
454
38 (4x AA 16x AF)
116.01 (4x AA 16x AF)
291 (4x AA 16x AF)
.
60
127.86
286
39 (4x AA 16x AF)
79.21 (4x AA 16x AF)
182 (4x AA 16x AF)
.
43
116.4
285
31 (4x AA 16x AF)
77.35 (4x AA 16x AF)
196 (4x AA 16x AF)
F.E.A.R - Performance Test, 1280 x 960
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
75
173.58
431
57 (4x AA 16x AF)
119.24 (4x AA 16x AF)
299 (4x AA 16x AF)
.
71
175.28
454
38 (4x AA 16x AF)
116.01 (4x AA 16x AF)
291 (4x AA 16x AF)
.
60
127.86
286
39 (4x AA 16x AF)
79.21 (4x AA 16x AF)
182 (4x AA 16x AF)
.
35
87.67
202
25 (4x AA 16x AF)
56.28 (4x AA 16x AF)
135 (4x AA 16x AF)
F.E.A.R - Performance Test. 1600 x 1200
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
66
108.88
214
30 (4x AA 16x AF)
60.6 (4x AA 16x AF)
135 (4x AA 16x AF)
.
56
106
221
23 (4x AA 16x AF)
45.36 (4x AA 16x AF)
112 (4x AA 16x AF)
.
39
69.51
138
18 (4x AA 16x AF)
40.16 (4x AA 16x AF)
87 (4x AA 16x AF)
.
26
61.65
136
18 (4x AA 16x AF)
36.73 (4x AA 16x AF)
18 (4x AA 16x AF)

Of course, the GeForce 8800GTS 640 and 320 MB, with more shader processing power than the Radeon X1950 Pro and GeForce 7950GT were able to offer higher minimum and average frame rates. Those of you who are finicky with minimum frame rates, like we do, should probably stick with the GeForce 8800GTS 640 MB. The card consistently offer higher minimum frame rates throughout the benchmark at 1024 x 768 and 1280 x 960 with AA and AF. At 1280 x 960 with AA and AF enabled, you can expect 46 to 50 percent higher frame rates overall, but only the GeForce 8800GTS 640 MB can offer 50 percent higher minimum fps compared to the GeForce 7950GT.

You may think the GeForce 8800GTS 640 MB still has high enough frame rates at 1600 x 1200 with AA / AF enabled, but we really don't recommend the card for that high a resolution. You'll want to have minimum frame rates no lower than 35 to 40 fps if you want to enjoy a real fluid gameplay experience in FEAR. In that respect, while both GeForce 8800GTS are faster than the Radeon X1950 Pro and GeForce 7950GT, we still think it's mainly still a card built for 1280 x 1024 or in this case 1280 x 960 with AA and AF enabled, at least with this game. So, there's actually very little going for it if you already using the GeForce 7950GT. It offers just about the same minimum fps, though it has slightly lower average fps.

The GeForce 8800GTS is definitely fast, however we can still see the penalty for AA is still pretty high. About 40 percent if you look at the results with and without AA and AF at 1600 x 1200.

F.E.A.R - Performance Test, 1024 x 768
GeForce 8800GTS 640
GeForce 8800GTS 320
57 (4x AA 16x AF)
119.24 (4x AA 16x AF)
299 (4x AA 16x AF)
47 (8x CSAA 16x AF)
105.04 (8x CSAA 16x AF)
261 (8x CSAA 16x AF)
37 (8xQ AA 16x AF)
80.77 (8xQ AA 16x AF)
215 (8xQ AA 16x AF)
42 (16x CSAA 16x AF)
105.6 (16x CSAA 16x AF)
257 (16x CSAA 16x AF)
22 (16xQ AA 16x AF)
59.41 (16xQ AA 16x AF)
155 (16xQ AA 16x AF)
.
38 (4x AA 16x AF)
116.01 (4x AA 16x AF)
291 (4x AA 16x AF)
33 (8x CSAA 16x AF)
97.93 (8x CSAA 16x AF)
249 (8x CSAA 16x AF)
34 (8xQ AA 16x AF)
76.38 (8xQ AA 16x AF)
213 (8xQ AA 16x AF)
31 (16x CSAA 16x AF)
102.69 (16x CSAA 16x AF)
264 (16x CSAA 16x AF)
23 (16xQ AA 16x AF)
54.96 (16xQ AA 16x AF)
150 (16xQ AA 16x AF)
F.E.A.R - Performance Test, 1280 x 960
GeForce 8800GTS 640
GeForce 8800GTS 320
33 (4x AA 16x AF)
84.13 (4x AA 16x AF)
199 (4x AA 16x AF)
30 (8x CSAA 16x AF)
72.27 (8x CSAA 16x AF)
178 (8x CSAA 16x AF)
23 (8xQ AA 16x AF)
56.19 (8xQ AA 16x AF)
144 (8xQ AA 16x AF)
36 (16x CSAA 16x AF)
72.77 (16x CSAA 16x AF)
175 (16x CSAA 16x AF)
21 (16xQ AA 16x AF)
41.1 (16xQ AA 16x AF)
103 (16xQ AA 16x AF)
.
39 (4x AA 16x AF)
80.11 (4x AA 16x AF)
197 (4x AA 16x AF)
24 (8x CSAA 16x AF)
66.26 (8x CSAA 16x AF)
169 (8x CSAA 16x AF)
25 (8xQ AA 16x AF)
50.99 (8xQ AA 16x AF)
141 (8xQ AA 16x AF)
29 (16x CSAA 16x AF)
70.57 (16x CSAA 16x AF)
178 (16x CSAA 16x AF)
17 (16xQ AA 16x AF)
31.76 (16xQ AA 16x AF)
70 (16xQ AA 16x AF)
F.E.A.R - Performance Test. 1600 x 1200
GeForce 8800GTS 640
GeForce 8800GTS 320
30 (4x AA 16x AF)
60.6 (4x AA 16x AF)
135 (4x AA 16x AF)
26 (8x CSAA 16x AF)
51.53 (8x CSAA 16x AF)
119 (8x CSAA 16x AF)
17 (8xQ AA 16x AF)
36.68 (8xQ AA 16x AF)
93 (8xQ AA 16x AF)
23 (16x CSAA 16x AF)
50.98 (16x CSAA 16x AF)
118 (16x CSAA 16x AF)
13 (16xQ AA 16x AF)
26.05 (16xQ AA 16x AF)
63 (16xQ AA 16x AF)
.
23 (4x AA 16x AF)
45.36 (4x AA 16x AF)
112 (4x AA 16x AF)
21 (8x CSAA 16x AF)
38.27 (8x CSAA 16x AF)
94 (8x CSAA 16x AF)
13 (8xQ AA 16x AF)
28.66 (8xQ AA 16x AF)
69 (8xQ AA 16x AF)
22 (16x CSAA 16x AF)
41.15 (16x CSAA 16x AF)
101 (16x CSAA 16x AF)
11 (16xQ AA 16x AF)
20.97 (16xQ AA 16x AF)
43 (16xQ AA 16x AF)

Usually, we have our fps progress graph where we look at how frame rates progress through the whole benchmark on each card. We opt no to do that today but rather see how much penalty we can expect when we push both GeForce 8800GTS card with higher modes of anti aliasing - 8xQ, 8x CSAA, 16xQ and 16x CSAA.

The trend seems constant - 8xQ with 8 color / Z / stencil samples is 'heavier' than 8x CSAA with just 4 color / Z / stencil samples. The good news is that there seem to be little performance difference between 8x and 16x CSAA. So, which modes are fast enough for fluid gameplay? Seeing we want at least 35 to 40 minimum fps, we recommend using 1024 x 768 with all these modes (except for 4x MSAA). Minimum frame rates seem to fluctuate between 30 to 35 fps at 1280 x 960 with the GeForce 8800GTS 640 MB and lower with the 320 MB version.

One more thing to note is that the added memory capacity of the GeForce 8800GTS 640 MB seems to have a positive effect on performance, even more so if you plan to use anti aliasing modes higher than 4x MSAA. The card is generally faster than its 320 MB sibling. Of course, we'll have to look at other benchmarks just to be sure.

Quake 4 - Data Processing Terminal, 1024 x 768
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
145
253.22
323
111 (4x AA 16x AF)
186.11 (4x AA 16x AF)
259 (4x AA 16x AF)
.
145
249.42
312
106 (4x AA 16x AF)
181.74 (4x AA 16x AF)
247 (4x AA 16x AF)
.
140
210.52
271
76 (4x AA 16x AF)
129.82 (4x AA 16x AF)
197 (4x AA 16x AF)
.
97
163.88
233
56 (4x AA 16x AF)
105.17 (4x AA 16x AF)
166 (4x AA 16x AF)
Quake 4 - Data Processing Terminal, 1280 x 1024
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
129
214.09
295
64 (4x AA 16x AF)
123.59 (4x AA 16x AF)
191 (4x AA 16x AF)
.
129
213.67
267
63 (4x AA 16x AF)
118.81 (4x AA 16x AF)
187 (4x AA 16x AF)
.
96
158.78
221
49 (4x AA 16x AF)
87.81 (4x AA 16x AF)
150 (4x AA 16x AF)
.
68
119.74
186
39 (4x AA 16x AF)
73.6 (4x AA 16x AF)
137 (4x AA 16x AF)
Quake 4 - Data Processing Terminal. 1600 x 1200
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
102
177.33
239
46 (4x AA 16x AF)
92.05 (4x AA 16x AF)
158 (4x AA 16x AF)
.
98
170.53
234
44 (4x AA 16x AF)
86.98 (4x AA 16x AF)
153 (4x AA 16x AF)
.
73
125.01
181
38 (4x AA 16x AF)
67.88 (4x AA 16x AF)
130 (4x AA 16x AF)
.
48
87.4
153
28 (4x AA 16x AF)
54.99 (4x AA 16x AF)
114 (4x AA 16x AF)

Basically, all these cards except the Radeon X1950 Pro have more than enough processing power to push high frame rates all the way up to 1600 x 1200 with AA and AF enabled. Of course, the GeForce 8800GTS 640 and 320 MB has a comfortable lead over the GeForce 7950GT and Radeon X1950 Pro. The performance difference came close to 40 percent for the GeForce 8800GTS 640 MB and 30 percent for the GeForce 8800GTS 320 MB when compared to the GeForce 7950GT at 1600 x 1200 with AA and AF enabled. The Radeon X1950 Pro with its OpenGL handicap trails behind.

Again, we're seeing the added memory capacity on the GeForce 8800GTS 640 MB coming into play. So, if you're planning to use ultra high resolutions, the GeForce 8800GTS 640 MB is a safer bet. We're also seeing about the same performance penalty with enabling AA - 40 to 50 percent at 1280 x 1024 and 1600 x 1200.

Quake 4 - Data Processing Terminal, 1024 x 768
GeForce 8800GTS 640
GeForce 8800GTS 320
111 (4x AA 16x AF)
186.11 (4x AA 16x AF)
259 (4x AA 16x AF)
93 (8x CSAA 16x AF)
166 (8x CSAA 16x AF)
244 (8x CSAA 16x AF)
72 (8xQ AA 16x AF)
126.09 (8xQ AA 16x AF)
195 (8xQ AA 16x AF)
91 (16x CSAA 16x AF)
164.85 (16x CSAA 16x AF)
243 (16x CSAA 16x AF)
57 (16xQ AA 16x AF)
105.02 (16xQ AA 16x AF)
170 (16xQ AA 16x AF)
.
106 (4x AA 16x AF)
181.74 (4x AA 16x AF)
247 (4x AA 16x AF)
88 (8x CSAA 16x AF)
159.55 (8x CSAA 16x AF)
230 (8x CSAA 16x AF)
69 (8xQ AA 16x AF)
121.06 (8xQ AA 16x AF)
188 (8xQ AA 16x AF)
86 (16x CSAA 16x AF)
157.75 (16x CSAA 16x AF)
230 (16x CSAA 16x AF)
56 (16xQ AA 16x AF)
101.36 (16xQ AA 16x AF)
165 (16xQ AA 16x AF)
Quake 4 - Data Processing Terminal, 1280 x 1024
GeForce 8800GTS 640
GeForce 8800GTS 320
64 (4x AA 16x AF)
123.59 (4x AA 16x AF)
191 (4x AA 16x AF)
58 (8x CSAA 16x AF)
110.35 (8x CSAA 16x AF)
182 (8x CSAA 16x AF)
44 (8xQ AA 16x AF)
81.49 (8xQ AA 16x AF)
143 (8xQ AA 16x AF)
57 (16x CSAA 16x AF)
108.63 (16x CSAA 16x AF)
171 (16x CSAA 16x AF)
36 (16xQ AA 16x AF)
68.25 (16xQ AA 16x AF)
120 (16xQ AA 16x AF)
.
63 (4x AA 16x AF)
118.81 (4x AA 16x AF)
187 (4x AA 16x AF)
57 (8x CSAA 16x AF)
108 (8x CSAA 16x AF)
179 (8x CSAA 16x AF)
42 (8xQ AA 16x AF)
78.7 (8xQ AA 16x AF)
142 (8xQ AA 16x AF)
49 (16x CSAA 16x AF)
106.81 (16x CSAA 16x AF)
171 (16x CSAA 16x AF)
35 (16xQ AA 16x AF)
65.38 (16xQ AA 16x AF)
114 (16xQ AA 16x AF)
Quake 4 - Data Processing Terminal. 1600 x 1200
GeForce 8800GTS 640
GeForce 8800GTS 320
46 (4x AA 16x AF)
92.05 (4x AA 16x AF)
158 (4x AA 16x AF)
37 (8x CSAA 16x AF)
82.57 (8x CSAA 16x AF)
141 (8x CSAA 16x AF)
31 (8xQ AA 16x AF)
58 (8xQ AA 16x AF)
116 (8xQ AA 16x AF)
41 (16x CSAA 16x AF)
82.04 (16x CSAA 16x AF)
143 (16x CSAA 16x AF)
24 (16xQ AA 16x AF)
45.35 (16xQ AA 16x AF)
80 (16xQ AA 16x AF)
.
44 (4x AA 16x AF)
86.98 (4x AA 16x AF)
153 (4x AA 16x AF)
42 (8x CSAA 16x AF)
81.19 (8x CSAA 16x AF)
139 (8x CSAA 16x AF)
31 (8xQ AA 16x AF)
56.77 (8xQ AA 16x AF)
112 (8xQ AA 16x AF)
41 (16x CSAA 16x AF)
80.35 (16x CSAA 16x AF)
137 (16x CSAA 16x AF)
23 (16xQ AA 16x AF)
43.85 (16xQ AA 16x AF)
76 (16xQ AA 16x AF)

Quake 4 is more forgiving than FEAR. You can get away with frame rates closer to 30 fps in this game and as the results suggest, you can get away with 8xQ anti aliasing with both the GeForce 8800GTS 640 and 320 MB at 1600 x 1200 with AA and AF enabled. Those of you who prefer 1280 x 1024, can opt to use a 16xQ if you want.

The trends we saw with FEAR persist - 8x and 16x CSAA seems to have about the same performance penalty and lighter than 'true' 8xQ MSAA. Again, of the two cards, you can get slightly higher frame rates, both average and minimum fps, with the GeForce 8800GTS 640 MB. However, the difference is not as big as we saw in FEAR.

Serious Sam II - Greendale, 1024 x 768
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
90
110.76
146
87 (4x AA 16x AF)
102.56 (4x AA 16x AF)
129 (4x AA 16x AF)
.
94
114.9
154
90 (4x AA 16x AF)
103.04 (4x AA 16x AF)
128 (4x AA 16x AF)
.
87
102.67
129
61 (4x AA 16x AF)
69.38 (4x AA 16x AF)
84 (4x AA 16x AF)
.
58
84.01
112
50 (4x AA 16x AF)
68.27 (4x AA 16x AF)
85 (4x AA 16x AF)
Serious Sam II - Greendale, 1280 x 960
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
90
108.38
141
70 (4x AA 16x AF)
81.16 (4x AA 16x AF)
105 (4x AA 16x AF)
.
94
110.86
144
68 (4x AA 16x AF)
79.42 (4x AA 16x AF)
103 (4x AA 16x AF)
.
72
85.48
109
46 (4x AA 16x AF)
51.7 (4x AA 16x AF)
64 (4x AA 16x AF)
.
55
74.33
93
41 (4x AA 16x AF)
54.99 (4x AA 16x AF)
69 (4x AA 16x AF)
Serious Sam II - Greendale. 1600 x 1200
GeForce 8800GTS 640
GeForce 8800GTS 320
GeForce 7950GT
Radeon X1950 Pro
82
93.11
116
51 (4x AA 16x AF)
59.62 (4x AA 16x AF)
78 (4x AA 16x AF)
.
82
94.13
116
50 (4x AA 16x AF)
58.43 (4x AA 16x AF)
76 (4x AA 16x AF)
.
55
65.28
85
32 (4x AA 16x AF)
36.58 (4x AA 16x AF)
46 (4x AA 16x AF)
.
44
57.74
70
31 (4x AA 16x AF)
41.22 (4x AA 16x AF)
52 (4x AA 16x AF)

Performance with and without AA and AF at 1024 x 768 are pretty close to each other with both GeForce 8800GTS - clearly a sign we're system limited. We need to use resolutions higher than 1280 x 960 or turn AA and AF on to see these cards stretch their legs. At 1600 x 1200 with AA and AF, both GeForce 8800GTS cards are around 60 percent faster than the GeForce 7950GT and 40 percent faster than the Radeon X1950 Pro.

Looking at the nominal frame rates, there's no doubt you can play Serious Sam II with all the settings maxed out at 1600 x 1200 with the GeForce 8800GTS, be it the 640 MB or 320 MB version. Unlike in F.E.A.R and Quake 4, we do not see any difference between the two versions here. You can try to play at 1600 x 1200 with the GeForce 7950GT and Radeon X1950 Pro, but undoubtedly you'll get a much more fluid gameplay experience with the GeForce 8800GTS.

Performance penalty incurred with AA enabled on the GeForce 8800GTS in this game is lower than it was with F.E.A.R. and Quake - slightly lower than 40 percent. It's still a big number though - overall the GeForce 8800GTS seems to continue the tradition NVIDIA cards have - 40 to 50 percent performance penalty with AA.

Serious Sam II - Greendale, 1024 x 768
GeForce 8800GTS 640
GeForce 8800GTS 320
87 (4x AA 16x AF)
102.56 (4x AA 16x AF)