[nextpage title=”How We Tested”]

During our benchmarking sessions, we used the configuration listed below. Between our benchmarking sessions the only variable was the video card being tested.

Hardware Configuration

• CPU: Core 2 Extreme QX9770 (3.2 GHz, 1,600 MHz FSB, 12 MB L2 memory cache).
• Motherboard: EVGA nForce 790i Ultra SLI (P05 BIOS)
• Memories: Crucial Ballistix PC3-16000 2 GB kit (BL2KIT12864BE2009), running at 2,000 MHz with 9-9-9-28 timings.
• Hard disk drive: Western Digital VelociRaptor WD3000GLFS (300 GB, SATA-300, 10,000 rpm, 16 MB cache).
• Video monitor: Samsung SyncMaster 305T (30” LCD, 2560×1600).
• Power supply: OCZ EliteXStream 1,000 W.
• CPU Cooler: Thermaltake TMG i1
• Optical Drive: LG GSA-H54N
• Desktop video resolution: 2560×1600 @ 60 Hz

Software Configuration

• Windows Vista Ultimate 32-bit
• Service Pack 1

Driver Versions

• nForce driver version: 15.17
• AMD/ATI video driver version: Catalyst 8.5
• AMD/ATI video driver version: Catalyst 8.6 + hotfix (8.501.1.0, 6/21/2008) (Radeon HD 4850, HD 4870)
• NVIDIA video driver version: 175.16
• NVIDIA video driver version: 177.34 (GeForce GTX 260, GTX 280)

Software Used

• 3DMark06 Professional 1.1.0 + October 2007 Hotfix
• 3DMark Vantage Professional 1.0.1
• Call of Duty 4 – Patch 1.6
• Crysis – Patch 1.2.1 + HardwareOC Crysis Benchmark Tool 1.3.0.0
• Half-Life 2: Episode Two – Patch June 9^th 2008 + HardwareOC Half-Life 2 Episode Two Benchmark Tool 1.2.0.0
• Quake 4 – Patch 1.4.2 
• Unreal Tournament 3 – Patch 1.2 + HardwareOC UT3 Benchmark Tool 1.2.0.0

Resolutions and Image Quality Settings

Since we were comparing very high-end video cards, we ran all our tests under three 16:10 widescreen high resolutions: 1680×1050, 1920×1200, and 2560×1600. We always tried to run the programs and games in two scenarios for each resolution, one with low image quality settings and then maxing out the image quality
settings. The exact configuration we used will be described together with the results of each individual test.

Error Margin

We adopted a 3% error margin; thus, differences below 3% cannot be considered relevant. In other words, products with a performance difference below 3% should be considered as having similar performance.

[nextpage title=”3DMark06 Professional”]

3DMark06 measures Shader 3.0 (i.e., DirectX 9.0c) performance. We run this software under three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600, first with no image quality enhancements enabled – results we call “low” on the charts and tables below –, then setting 4x anti-aliasing and 16x anisotropic filtering. See the results below.

[nextpage title=”3DMark Vantage Professional”]

3DMark Vantage is the latest addition to the 3DMark series, measuring Shader 4.0 (i.e., DirectX 10) performance and supporting PhysX, a programming interface developed by Ageia (now part of NVIDIA) to transfer physics calculations from the system CPU to the video card GPU in order to increase performance. Mechanical physics is the basis for calculations about the interaction of objects. For example, if you shoot, wh
at exactly will happen to the object when the bullet hits it? Will it break? Will it move? Will the bullet bounce back? Notice that we didn’t upgrade the PhysX to the latest version, which would make the physics calculations for CPU Test 2 to be made by the GPU instead of the CPU on NVIDIA video cards (since we aren’t considering CPU or 3DMark scores this change wouldn’t produce any increase in our results anyway).

We ran this program at three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600. First we used the “Performance” profile, and then we used the “Extreme” profile (basically enabling anti-aliasing at 4x, anisotropic filtering at 16x, and putting all detail settings at their maximum or “extreme” value. The combination of 2560×1600 resolution with extreme settings didn’t produce reliable results according to the program, so we aren’t going to add them here. The results being compared are the “GPU Scores” achieved by each video card.

[nextpage title=”Call of Duty 4″]

Call of Duty 4 is a DirectX 9 game implementing high-dynamic range (HDR) and its own physics engine, which is used to calculate how objects interact. For example, if you shoot, what exactly will happen to the object when the bullet hits it? Will it break? Will it move? Will the bullet bounce back? It gives a more realistic experience to the user.

We ran this program at three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600, maxing out all image quality controls (i.e., everything was put on the maximum values on the Graphics and Texture menus). We used the game internal benchmarking feature, running a demo provided by NVIDIA called “wetwork.” We are putting this demo for downloading here if you want to run your own benchmarks. The game was updated to version 1.6.

[nextpage title=”Crysis”]

Crysis is a very heavy DirectX 10 game. We updated this game to version 1.2.1 and used the HOC Crysis Benchmarking Utility to help us collecting data. Since we don’t think the default demo based on the island map stresses the video card the way we want, we used the HOC core demo available with the abovementioned utility. We ran this demo under three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600, first with image quality set to “low” and then with image quality set to “high.” Since all video cards achieved a number of frames per second below 10 at 2560×1600 with image details set to “high,” we are not including this test as the results aren’t reliable. We ran each test twice and discarded the first result, as usually the first run achieves a lower score compared to the subsequent runs since the game loses time loading files. The results are below, in frames per second (FPS).

29

[nextpage title=”Unreal Tournament 3″]

Unreal Tournament 3 is the latest installment from this famous first person shooter franchise, supporting DirectX 10 graphics when installed on Windows Vista with a DX10 compatible card. We upgraded Unreal Tournament 3 to version 1.2 and benchmarked it with the help of HOC UT3 benchmarking utility using the “Containment” demo, maxing out image quality settings (image quality at “high” and anisotropic filtering at x16). It is important to note that we haven’t applied the PhysX mod to this game, which would transfer PhysX processing from the CPU to the GPU on NVIDIA cards. The results are below, in frames per second (FPS).

[nextpage title=”Half-Life 2: Episode Two”]

Half-Life 2 is a popular franchise and we benchmark the video cards using Episode Two with the aid of HOC Half-Life 2 Episode Two benchmarking utility using the “HOC Demo 1” provided by this program. We ran the game in three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600, under two scenarios. First with quality set to maximum, bilinear filtering and anti-aliasing set to x0. This configuration we are calling “low” on the charts and tables below. Then we maxed out image quality settings, enabling x16 anisotropic filtering and 16xQCS anti-aliasing. This configuration we are calling “high” on our charts and tables. We updated the game up to the June 9^th 2008 patch.

< tr>

[nextpage title=”Quake 4″]

We upgraded Quake 4 to version 1.4.2 and ran its multiplayer demo id_perftest with SMP option enabled (which allows Quake 4 to recognize and use more than one CPU), under the same three 16:10 widescreen resolutions, 1680×1050, 1920×1200, and 2560×1600, first with image quality settings configured at “low” and then with image quality settings configured at “ultra.” You can check the results below, given in frames per second.

[nextpage title=”Conclusions”]

With so many results presented in the previous page, we will make a summary of how is XFX GeForce GTX 260 640M XXX against the standard GeForce GTX 260, GeForce GTX 280 (as this XFX model runs at a higher clock rate) and Sapphire HD 4870, which is today the main competitor to the standard GTX 260.

XFX GeForce GTX 260 640M XXX was up to 19% faster than the standard GeForce GTX 260, depending on the program and video configuration. You will achieve, on average, a 10% performance increase with this overclocked model from XFX, which is quite interesting, as it costs around 10% more than the standard GTX 260. Paying ten percent more to have ten percent more performance looks very attractive, especially when you think that price and performance usually don’t increase at the same proportion.

GeForce GTX 280, on the other hand, was up to 20% faster than this overclocked model from XFX, but on some games like Unreal Tournament 3 and Half-Life Two: Episode Two the reviewed card was in fact faster than GTX 280. This overclocked GTX 260 from XFX runs with clocks higher than the ones used on GTX 280, so on games that clock plays a more important role the reviewed card will be faster. On the other hand, on games that more memory bandwidth and more streaming processors play a more important role (Crysis and Call of Duty 4, for example), GTX 280 will be faster.

The reviewed video card from XFX was up to 22% faster than Radeon HD 4870, depending on the video configuration used. In some scenarios (some configurations of Crysis, Unreal Tournament 3, Half-Life 2: Episode 2 and Quake 4) both card achieved similar performance, with HD 4870 being faster on some configurations of Quake 4 and Half-Life 2: Episode Two.

Here are some thoughts about XFX GeForce GTX 260 640M XXX, keeping in mind that this video card is targeted to users that have USD 300 to spend on a video card.

If you are planning on buying Call of Duty 4, then this video card is a terrific option, as it comes with this game. Since this game costs around USD 50 you are in fact paying USD 280 for this video card (USD 330 – USD 50).

If you had already decided to buy a GeForce GTX 260, then this model from XFX can be an interesting option. For 10% increase in price you get a 10% increase in performance, on average. Not a bad deal.

But if you are a looking for a high-end video card with a good cost/benefit ratio and don’t give a damn about Call of Duty 4, then you will probably be better off buying a Radeon HD 4870.