Test setup

All of today's testing has been run on the following:

- Intel Core 2 Duo E6600
- 2GB Corsair PC6400 DDR2 RAM
- MSI 975X Platinum PowerUP Edition (Socket LGA 775, PCI Express)
- 250GB Western Digital Caviar SE16 hard drive
- Pioneer 16x DVD-ROM
- NVIDIA GeForce 8800 GTS 640MB
- Sapphire Radeon HD 2900 XT 512MB
- 1000W Thermaltake Toughpower power supply
- Windows Vista Home Premium (64-bit Edition)

The following drivers were used:

- NVIDIA ForceWare 158.18 was used on the GeForce 8800 GTS
- ATI Catalyst 7.5 was used on the Radeon HD 2900 XT

The drivers High Quality texture filtering mode was enabled on both boards for all real-world game testing.

Benchmarks used

Below is a list of all the tests and applications used for benchmarking:

- 3DMark 06 (Version 1.1.0)
Overall score, fillrate, perlin noise and batch size tests, 1600x1280, AA/AF disabled
- ShaderMark 2.1 (Build 129)
All tests, 1280x768, AA/AF disabled
- Half-Life 2: Episode One
Elite Bastards custom timedemo, application AA/AF
- Prey (Version 1.3)
Elite Bastards custom timedemo, application AA/AF
- Elder Scrolls IV: Oblivion (Version 1.1)
FRAPS, application AA, control panel AF
- Company of Heroes (Version 1.70)
Built-in timedemo, application AA, control panel AF
- F.E.A.R. (Patch 1.08)
Built-in timedemo, application AA/AF
- S.T.A.L.K.E.R.: Shadow of Chernobyl
FRAPS, application AA/AF
- Need For Speed: Carbon (v1.3)
FRAPS, application AA/AF
- Lost Planet: Extreme Condition demo (DirectX 10)
Built-in timedemo, application AA/AF

Synthetic benchmarks

The full focus of today's review will be how Sapphire's Radeon HD 2900 XT compares to a reference clocked GeForce 8800 GTS 640MB board - It's closest competitor at this time based on price point.  We've used ATI's first WHQL driver for this board for our testing, putting it through its paces using Catalyst 7.5 throughout.

3DMark06

3DMark06 is our first port of call, beginning with a glance at both Shader Model 2.0 and 3.0 performance using the benchmark's default settings.

Starting out with 3DMark06's Shader Model 2.0 tests, the Radeon HD 2900 XT jumps into an early lead over its major rival, outperforming the 640MB GeForce 8800 GTS by a little over 10%. 

Even more impressive is the Radeon HD 2900 XT's Shader Model 3.0 and High Dynamic Range rendering performance, where it beats out the GeForce 8800 GTS by not far short of 35%.

3DMark06's theoretical fill rate testing shows off the Radeon HD 2900 XT's advantage when it comes to single texturing, whereas multi-texturing gives basically very little difference between the two boards.

The math intensive Perlin Noise shader test is an interesting synthetic example of a situation featuring a very ALU heavy workload - This kind of scenario seems to suit the Radeon HD 290 XT down to the ground, where it offers up a huge 73% advantage over the NVIDIA board. 

A mixture of the higher core clock speed and larger number of shader units available to the Radeon HD 2900 XT combine to give it a lead in both of 3DMark06's synthetic vertex shading tests, with the complex vertex shader scenario in particular giving the ATI board close to a 75% lead once more.

3DMark06's 'Shader Particles' test examines the performance of Vertex Texture Fetch functionality - While NVIDIA boards have supported this ability since the GeForce 6 series, the Radeon HD 2000 series is the first time this capability has been available on ATI hardware.  Judging by this theoretical example, ATI also appears to have the upper hand in this discipline as well, leading by 33% from the GeForce 8800 GTS.

Finally, this benchmarking suite's synthetic batch size testing investigates how performance scales from handling small batches (which is often driver-limited) up to much larger sized batches of pixels.  Our results here are interesting, seeing the GeForce 8800 GTS handle small batches of pixels slightly more efficiently, but dropping some way behind when dealing with large batch sizes.

ShaderMark

Next up, we focus completely on this board's pixel shading, and in particular Pixel Shader 3.0 performance, courtesy of ShaderMark 2.1.  Thankfully, this benchmark works perfectly under Windows Vista provided 'Run as Administrator' is selected to run the application.

Overall the Radeon HD 2900 XT wins out here, although there are a handful of tests that give victory to the NVIDIA board.  Most interesting here are the tests which involve High Dynamic Range rendering, which serve to show off the massive improvements made to R600 in this discipline, with the Sapphire board wiping the floor with the GeForce 8800 GTS.