Tech-Hounds.com

A Direct Comparison of Radeon X700 Pro and GeForce 6600GT

PCI Express x16 graphics cards
Chipset Radeon X700 Pro 256 MB, 128 bit, GeForce 6600 128 MB, 128 bit

Last week, we featured two round up articles – the Radeon X700 Pro and the GeForce 6600GT round up. Some of you may wonder why we didn't directly compare both cards. This article is the answer. You might be surprise by the outcome: while NVIDIA does have the faster hardware, we still recommend Radeon X700 Pro over the GeForce 6600GT. Now, before you start flaming and shouting 'fanboy', permit us to present our case.



There are several reasons why we put off doing direct comparison between Radeon X700 Pro and GeForce 6600GT. During testing for the round ups, we used a slightly older driver (Catalyst 5.4 and Forceware 71.22, if you haven't noticed). Since then, we've upgraded to the latest final release from both ATI and NVIDIA, which is Catalyst 5.7 and (still) Forceware 77.72. Another reason was to investigate a driver bug, which to our dismay is still present on the Forceware 77.72. This bug causes very slow performance in F1 Career Challenge when shadows are used at full detail. Lowering shadow detail results in shadows not rendering at all. This has prompted us to perform additional testing with old drivers, the Forceware 66.81, which debut around the same time as the GeForce 6600GT. To make the comparison fair, we also perform additional testing with Catalyst 4.12. These additional testing leads us to another discovery: there is a rendering 'bug' with Splinter Cell present in Forceware 71.22 and 77.72. The 'bug' affects how the ocean's ripple are portrayed in the Caspian Oil Refinery demo, which are done with pixel shaders. As you can see from the screenshots, this really affects the overall image quality. This bug is not present in Forceware 66.81.



Of course, this complicates things. It reeks of 'shader replacement', a practice not unfamiliar with NVIDIA. This 'bug' or whatever you want to call it may affect benchmark results. We say 'may', because we have no idea just how much (if any) influence these bug have on the results: in general, Forceware 77.72 (and 71.22) is slower by 1 fps than the 66.81. So, we conduct even more additional testing with the included official demo, the 1_1_2TbilisiDemo.bin. This demo performs much alike the Beyond3D demo, but lacks the ocean pixel shader. Results are similar (slower or faster by 1 fps).

Optimizations issues aside, these 'bugs' complicate evaluation: do we evaluate the cards using old or newer drivers? Is it fair to use older drivers in the evaluation because the GeForce 6600GT has 'bugs' which the Radeon X700 does not have? To answer all these questions (and more), we've decided to conduct test with both old and new drivers for both cards. This way users can still see the performance of the cards without the presence of bugs we have found. We also expanded the benchmarks to include more games, not just our usual graphics and system benchmarks but also other, older games we could get our hands on. A backward looking review, you might say. We're pretty sure this paints a more clear picture of what these cards offer not just for the FPS oriented crowd, but also appeals to the general gamer. We want to see if there are other 'bugs' present in Forceware 77.72. Of course, this will also let us evaluate driver progress from both NVIDIA and ATI. Except for the two games (Splinter Cell and F1 Career Challenge), just how much performance is lost or gained with a new driver in older games? The obligatory image test are also expanded to include older drivers, in fact it is this testing that showed us the Splinter Cell 'bug' in the first place. However, due to time constraints, we didn't perform exhaustive performance and image comparisons test for the additional games (system benchmarks and extra games). However, we do still take note the image quality on these extra games. Testing are limited to 1024 x 768, 32 bit with and without 4x AA and 16x AF. These settings are what users of both cards are generally using anyhow (and they perform less than stellar at higher resolutions). We didn't update the game, partly because we didn't have time and also because we want to see just how 'compatible' the cards are. So, any performance loss and image quality differences may be caused by outdated binaries.

Overview

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

ATI X700 Pro 256 MB
Core Clock : 425 MHz
Memory Clock : 433 MHz (866 MHz DDR3)
Memory Bus Width : 128 bit

NVIDIA GeForce 6600GT 128 MB DDR3
Core Clock : 500 MHz
Memory Clock : 500 MHz (1000 MHz DDR3)
Memory Bus Width : 128 bit

Image Quality Comparison

The cards we test are not reference cards, but they are based on the reference design. So for all intent and purposes, they represent Radeon X700 Pro and GeForce 6600GT quite well. However, some manufacturer may use a lower or higher clock, so these cards will probably perform slower or faster (but not by much).

The games we tested with:
Graphics benchmarks:
Call of Duty
Splinter Cell
F1 Career Challenge
Nascar Season 2003

System benchmarks:
Dungeon Siege
Lock On: Modern Air Combat
Full Spectrum Warrior

Extra benchmarks:
Need for Speed: Underground
Battlefield 1942
Star Trek: Bridge Commander
Serious Sam: Second Encounter

Our test setup
AMD Athlon 64 3000+ 'Newcastle' 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 5.7 & 4.12 reference driver
NVIDIA Forceware 77.72 & 66.81 reference driver
NVIDIA NForce system driver version 6.66
Creative SoundBlaster Live! 24 bit integrated (motherboard supplied driver)
DirectX 9.0c

Below you can find screenshots we've taken for image comparison purposes. The left image is without AA and AF, while the right image (if any) are with AA and AF. You could click on the images to see the full image (1024 x 768, 32 bit). These full images are quite big, so they may take a while to load. A note on Splinter Cell's images: since the game uses pixel shaders, the oceans doesn't look the same. Instead of looking at the ocean, try looking at Sam and the oil rig.

Catalyst 5.7 and 4.12

Call of Duty

Catalyst 4.12


Catalyst 5.7



No difference between the drivers for both default (Texture Preference: high quality/ MIP Detail Level: high quality) and 4x AA and 16x AF screenshots.

Splinter Cell

Catalyst 4.12



Catalyst 5.7



No difference between the drivers for both default (Texture Preference: high quality/ MIP Detail Level: high quality) and 16x AF screenshots.

F1 Career Challenge

Catalyst 4.12



Catalyst 5.7



Except for the sky, trees and tires there are no differences between the drivers for both default (Texture Preference: high quality/ MIP Detail Level: high quality) and 4x AA and 16x AF screenshots. The above variations are normal and doesn't affect benchmark results.

Nascar 2003

Catalyst 4.12



Catalyst 5.7


No difference between the drivers for both default (Texture Preference: high quality/ MIP Detail Level: high quality) and 16x AF screenshots.

Forceware 77.72 and 66.81

Call of Duty

Forceware 66.81



Forceware 77.72



No noticeable difference (when in motion) between the drivers for both default (Image Setting: Quality) and 4x AA and 16x AF screenshots. Naturally, there are difference between High Quality and Quality settings, but not between drivers.

Splinter Cell

Forceware 66.81



Forceware 77.72



Differences in ocean ripples (due to pixel shader 'bug')

F1 Career Challenge

Forceware 66.81



Forceware 77.72



Except for the sky, trees and tires there are no difference between the drivers for both default (Image Setting: Quality) and 4x AA and 16x AF screenshots. The above variations are normal and doesn't affect performance results. There are no noticeable differences between High Quality and Quality settings.

Nascar 2003

Forceware 66.81



Forceware 77.72



Differences can be seen on the side netting, logo on the door side of the car (left side). There are also differences in the fence and grandstand (right side) and in the mirror. Overall, the 66.81 screenshot has a sharper feel (fence and grandstand most probably not noticeable, but side netting, logo on door and cars are). On 4xAA and 16x AF screenshots, there's no noticeable differences, but again the 66.81 has a sharper feel.





Catalyst 5.7/4.12 and Forceware 66.81

Image quality is very much the same for Catalyst 4.12 and 5.7, but we're using the screenshots under 5.7 in comparison to Forceware 66.81 which lacks both bugs (image quality and performance), plus it has s sharper feel for Nascar 2003.

Call of Duty

Catalyst 5.7



Forceware 66.81



While there are differences between NVIDIA's default quality and ATI's High Quality image, they are not noticeable (especially in motion). The same with 4x AA and 16x AF screenshots.

Splinter Cell

Catalyst 5.7



Forceware 66.81



No difference between the drivers for both default (Texture Preference: high quality/ MIP Detail Level: high quality on Catalyst and Image Settings: Quality on Forceware) and 16x AF screenshots.

F1 Career Challenge

Catalyst 5.7



Forceware 66.81



Except for the sky, trees and tires there are no difference. In the 4x AA and 16x AF screenshots, it's hardly noticeable in motion but in still images the Forceware 66.81 is sharper. Catalyst 5.7 have a smoother image.

Nascar 2003

Catalyst 5.7



Forceware 66.81



A small, barely noticeable difference can be seen on the side car's front right tire. On 4xAA and 16x AF screenshots, there's no noticeable differences, but again the 66.81 has a sharper feel while Catalyst 5.7 maintains a smooth image.

So image quality wise, the Forceware 77.72 used by the GeForce 6600GT differs a lot with Catalyst 5.7 (and of course Catalyst 4.12) used by the Radeon X700 Pro or even Forceware 66.81. Noticeable differences in default image quality are present in Nascar 2003. In Nascar 2003, with 4x AA and 16x AF, these differences are not noticeable (Catalyst 5.7 slightly sharper compared to Forceware 77.72).

Performance

The results:

Catalyst 4.12 and 5.7

With Call of Duty, there are differences but hardly noticeable and certainly not significant (3 fps at most). We see a 1 fps difference in Splinter Cell and at most 4 fps difference without 4x AA and 16x AF in F1 Career Challenge. The same with Nascar 2003. Looks like ATI hit it off right at the start with Catalyst 4.12. Later driver versions are mainly focused at squashing bugs.

Forceware 66.81 and 77.72

Looks like even the 66.81 is not bug free, at least with Call of Duty. During testing, an anomaly occur at 1280 x 1024, 32 bit (dropping further than it should – below 100 fps). Restarting the game several times fixed the problem. For both 1024 x 768, 32 bit and 1600 x 1200, 32 bit, the results from both driver versions are more or less the same (at most 5 fps from a 140 fps result). The Caspian Oil Refinery test results only differs by 1 fps, confirmed by Tbilisi Police Precinct demo with the 66.81 being faster. The bug in Forceware 77.72 makes F1 Career Challenge definitely unplayable with 4x AA and 16x AF. But we could already see the influence even in without AA and AF in resolutions higher than 1024 x 768, 32 bit. For Nascar 2003, the 77.72 is slower except at 1600 x 1200, 4x AA and 16x AF. In this regard, the Forceware 66.81 is the better driver: its faster, has no bug in Splinter Cell and F1 Career Challenge plus it maintains image quality when used without AA and AF in Nascar 2003.

Catalyst 5.7 and 66.81/77.72

Image quality wise, this is the most ideal comparison between the Radeon X700 Pro and GeForce 6600GT under the circumstances. The score from Forceware 77.72 is included just as a reminder.

In Call of Duty, we're seeing a difference between 6 fps in 1024 x 768, 32 bit to 20 fps in 1600 x 1200, 32 bit which is quite significant. With 4x AA and 16x AF, the differences are 19 fps in 1024 x 768, 32 bit and 14 fps in 1600 x 1200, 32 bit. While these differences are significant, that doesn't mean the Radeon X700 is not suited for Call of Duty – it can still provide an average 108.8 fps in 1024 x 768, 32 bit and still playable at 62.4 fps in 1600 x 1200, 32 bit with 4x AA and 16x AF.

There's no doubt that Splinter Cell runs faster on the GeForce 6600GT. NVIDIA's hardware is just more efficient with rendering shadows under projector mode. We're seeing a difference of 13.68 average fps between the two. This difference can also be seen in the minimum fps – 9.88 fps. Unfortunately, that difference narrows at 1280 x 1024, 32 bit and 1600 x 1200, 32 bit. If you look at these results, this demo only gives playable results for both cards in 1024 x 768, 32 bit color – anything higher and you're dipping below 30 fps at some point. Please keep in mind these situations don't occur all the time, only on levels with lots of lights and shadows.

In F1 Career Challenge, the Radeon X700 Pro is faster without AA and AF up to 1280 x 1024, 32 bit. In 1600 x 1200, 32 bit the GeForce 6600GT has a faster average fps, but both cards have the same minimum fps. With 4xAA and 16xAF, the GeForce 6600GT takes the lead, right up to 1600 x 1200, 32 bit where both cards are barely passing the 30 fps mark. The results from Forceware 66.81 certainly proves that the GeForce 6600GT has what it takes for fluid gameplay with AA and AF. The Radeon X700 Pro is faster than GeForce 6600GT without AA and AF, so what happened here? The performance hit with AA and AF shouldn't be that big. Apparently, the answer lies in texture filtering. Changing the image quality slider in Forceware doesn't have any performance or quality effects, but doing the same thing with the Texture Preference slider in Catalyst does – performance wise! As far as we can tell, there's no difference in image quality even under High Performance mode. This is more of an acceptable compromise for users than not rendering shadows at all to get playable frame rates with Forceware 77.72. For this article, we're sticking to the default Texture Preference (High Quality).

For the most part, Nascar 2003 behaves the same way with both cards. We can see there's a performance difference, but they are hardly noticeable. Only on 1600 x 1200, 32 bit with 4x AA and 16xAF does the difference become noticeable (7 fps).

Combined with what we discover with our image comparison test, the picture is not very clear. Under Forceware 68.11, two games - Call of Duty and Splinter Cell – are faster with and without AA and AF on the GeForce 6600GT with no noticeable image quality differences. F1 Career Challenge is faster without AA and AF on the Radeon X700 Pro, but only through adjustment on the Texture Preference slider we see a closer gap (4 to 6 fps) between the two cards with AA and AF. On the other hand, Nascar is just the Radeon X700 Pro home ground: better image quality without AA and AF plus faster performance.

So what driver should we choose to evaluate the GeForce 6600GT? If we were to use the 66.81, the GeForce 6600GT is the faster card with comparable image quality. But if we use the 77.72, it is the slower card with 'lower' image quality. The Radeon X700 Pro maintains both performance and quality in both driver versions (Catalyst 4.12 and 5.7). Even with this limited, problematic and confusing benchmark results, you can see that overall the Radeon X700 Pro has a better combination of hardware and software. Performance between driver versions are consistent, even constant and this also ring true with its image quality. The GeForce 6600GT is the faster card here, probably due to its higher clock (a 75 MHz difference) and being more efficient in Splinter Cell's shadow rendering method. But the drivers have some 'bugs'. Let's put off making a conclusion for now. What about other games?

In Average fps	Catalyst 4.12 - Default	Catalyst 4.12 – 4x AA 16x AF	Catalyst 5.7 - Default	Catalyst 5.7 – 4x AA 16x AF	Forceware 66.81 - Default	Forceware 66.81 – 4x AA 16x AF	Forceware 77.72 - Default	Forceware 77.72 – 4x AA 16x AF
Lock On	57.33	45.94	57	46.4	54.46	46.63	56.37	48.86
FSW	65.85	50.02(AF only)	66.91	67.06(AF only)	70.52	68.94 (AF only)	63.72	57.43(AF only)
NFSU	64.8	65.03	66.56	64.64	62.27	62	64.72	64.05
BF1942	74.75	71.55	85.42	73.9	83.3	75.04	82.32	75.93
STBC	177.67	130.27	173.36	128.66	183.53	129.52	167.91	129.16
DS	107.72	100.8	107.26	101.19	103.24	95.73	104.1	99.3
SS:SE	78.8	65.2	78	67.8	96.6	86.9	80.7	56.2

Funny how even system limited benchmarks like Dungeon Siege and Lock On are generally faster on the Radeon X700 Pro (though hardly noticeable). Full Spectrum Warrior is faster with Forceware 66.81 than 77.72, just like Star Trek: Bridge Commander. Results from Battlefield: 1942 should not be taken 'as is', since these are not based on replays, but frame rates in gameplay with relatively the same elements. The same also applies to Need For Speed: Underground but this game is more predictable and as such repeated benchmark results are identical. All of the games benchmark results from both cards are very close to each other.

Yet, there is something interesting with Serious Sam: Second Encounter benchmark results. We chose the Valley of the Jaguar demo for this run, running it with settings applied by the addon script 'Extreme Quality'. Just look at the performance loss we get from moving to Forceware 77.72 – 16 to 30 fps! Another 'bug'? Honestly, we don't know – it could just be a bug in the game (we didn't patch the game to the latest patch – 1.5).

You might say that these benchmarks are system limited and thus results using a higher resolution should be viewed before making a decision. Unfortunately, its not always possible. You can't use a higher resolution than 1024 x 768 with Dungeon Siege, while with Lock On, you probably don't want to since it will be unplayable. Full Spectrum Warrior is limited to a maximum resolution of 1280 x 1024 as do Need for Speed: Underground. The other games can use resolutions up to 1600 x 1200, 32 bit. But look at the results: these are average fps and they're already bordering in 60 to 80 fps, minimum frame rates will be lower. We want to keep the minimum frame rates above 30 fps if possible. And if you're really want to play games at those resolutions, you're better off with a GeForce 6800 / 7800 or Radeon X800 / X850 with at least an Athlon 3500+, not the GeForce 6600GT and Radeon X700 Pro with an Athlon 3000+.

Here are some notes regarding image quality and graphical issues we saw:

• On Forceware 66.81 and 77.72, bloom in Full Spectrum Warrior is slightly off, resulting in white bars (the bloom effect) on the sides and bottom of the screen when you're looking at the sky. These are not present on Catalyst 4.12 and 5.7.
• Activating AA and / or AF in Battlefield: 1942 results in garbled text with both Forceware versions. None are present on Catalyst drivers.
• In Star Trek: Bridge Commander, we can see some artifacts on the ship's texture, with both Forceware versions. None are present on Catalyst drivers.

Its hard not to see a trend here: very strange since most of these games are based on DirectX 8 or 8.1 (with the exception of Serious Sam: Second Encounter). Again, all these bugs may be fixed by a binary update, but it is strange that the both Catalyst versions are having none of the problems. Since they're based on DirectX 8, NVIDIA shouldn't have this problem. Of these seven games, four of them experience minor improvements (hardly noticeable in gameplay) with Forceware 77.72. The three games that experience a performance drop is about 10 % slower without AA and AF. With AA and / or AF, these performance drop range from 16 to 30 % (Full Spectrum Warrior and Serious Sam: Second Encounter). With these in mind, there's no doubt while performance in these extra games are generally comparable between the GeForce 6600GT and Radeon X700 Pro, the image quality is overall better on the Radeon X700 Pro.

Of these seven extra games, only Full Spectrum Warrior and Serious Sam: Second Encounter are faster in Forceware 66.81 than either Catalyst (4.12 or 5.7) with both without and with 4xAA and 16xAF. The difference is only significant with Serious Sam: Second Encounter. Star Trek: Bridge Commander is faster without AA and AF, but it's only 10 fps out of 180 fps (not significant and certainly not noticeable). So, off all the game we test (11 in total), the GeForce 6600GT with Forceware 66.81 is only significantly faster in four games (Call of Duty, Splinter Cell, F1 Career Challenge and Serious Sam: Second Encounter). Four out of eleven is not a good enough number to declare a clear win for the GeForce 6600GT.

Heat

While we didn't have heat issues with either card (both are using reference cooling solutions), we'd like to share some information with you regarding heat and temperature. To see just how hot these cards can get, we continuously loop Doom 3's 'demo1' for 5 minutes, under High Quality settings (1024 x 768, no AA, application defined 8x AF). The temperature are viewed through Forceware default temperature monitoring and manufacturer's supplied utility. Here are the results:

	Core Idle	Core Full Load	Ambient
Radeon X700 Pro	33º Celsius	48º Celsius	43º Celsius
GeForce 6600GT	46º Celsius	62º Celsius	43º Celsius

The GeForce 6600GT is hot. So, if you're using the computer in a quite hot environment, better be prepared to buy a more effective cooling solution for the GeForce 6600GT. or at least have a better air circulation inside your case. Cool down period is relatively quick: one minute after full load with zero load (desktop).

Other Miscellaneous Information

There are several beta Forceware lying around and a final, WHQL qualified driver is expected in the coming months. Until they arrive, we don't know if they will fix the bugs present in the current Forceware 77.72. They might, or they might not.

Just like we stated before, we're aware that the GeForce 6600GT is the faster card (both in clock, bandwidth and benchmark results). But think for a minute: what card is useful without good drivers? These drivers are what influence their performance in games, not just faster clocks and larger bandwidth. With a 75 MHz difference, we don't want the same performance, we want significantly faster performance (above 10%)! But it looks like we're only getting it if the developers code specific support in their game (Splinter Cell and Call of Duty). If you're very focused on certain games like these, the GeForce 6600GT will provide faster performance at comparable image quality.

There is additional features of GeForce 6600GT such as Shader Model 3 and SLI support that are interesting, but not without some caveats. Up till now, the amount of games fully utilizing SM 3.0 capabilities are limited – FarCry and Splinter Cell: Chaos Theory comes to mind. There are performance gains with SM 3.0, but you have to ask yourself whether or not they're significant for you. Unfortunately, we didn't test the two games, but there are other sites that do. So, we advise you to look for benchmark results between the two cards in those games if SM 3.0 performance is important to you.

There's also high dynamic range (HDR) rendering, the latest wonder in real time 3D graphics made possible by SM 3.0. But look at the HDR benchmarks: if a GeForce 6800 Ultra is barely coping with 30 to 40 average fps, why would you think the GeForce 6600GT will be better off? The drivers certainly won't provide magical improvements. A GeForce 6600GT with HDR rendering is like the GeForceFX 5200 with DirectX 9: you can but trust us, you wouldn't want to.

SLI support will give you performance improvements, but not without caveats – right now, HDR is not possible on SLI setups. So, you're probably better off with a single GeForce 6800GT than two GeForce 6600GT (or the coming GeForce 7800GT).

There's also the lack of AA with HDR, so again we encounter the same problem with DirectX 8 bloom effects. This is a sad situation, gamers should never be faced with 'either / or' situation like this. With the GeForce 7800 series, you could probably suppress this by using a higher resolution, but that's not an option for the GeForce 6600GT (or even the 6800). Anti aliasing remains the more practical choice of image enhancement.

Conclusion:

For the less technically inclined gamer, incompatibilities and bugs with games old an new can be very frustrating. Even more so if the bugs are causing horrendous slowdown or worse, crashes. The Radeon X700 Pro is better suited for them and gamers looking for an all around graphics card. It simply works, the image quality is superb and its close to or just as fast as the GeForce 6600GT! Benchmark results with the exception of Call of Duty and Splinter Cell are very close if not identical between the Radeon X700 Pro and GeForce 6600GT. Of course, Catalyst is the better driver here.

The GeForce 6600GT is faster, but their drivers are not up to par with the Radeon X700 Pro's. Even more alarming is the fact that the 'bugs' affect pre-DirectX 9 games. Of course, you can use the older, more matured drivers (like we did with the Forceware 66.81), but this probably means forsaking fixes for newer games. New drivers are supposed to be better, if not faster. The current trend with Forceware doesn't seem to reflect this. ATI's Catalyst drivers are way better here: offering consistent and constant performance plus bug fixes. NVIDIA may be the 'golden standard' for developers, but they sure as hell not for gamers. Forceware developers, take a hint – fix the bugs!!!

Go to top