Test System Setup
The graphics boards tested for this article were installed in an Antec Sonata midtower case and the testbed built using the following components:
- AMD Athlon 64 FX-60 (2.6 GHz dual core)
- ASUS A8N SLI Deluxe motherboard (nForce4 chipset)
- 2 GB (2 x 1 GB) of Corsair DDR400 memory (2.5-3-3-6)
- Maxtor MaxLine III 250 GB SATA (16 MB buffer) hard drive
- Plextor PX-712S DVD drive
- Creative SoundBlaster Audigy 2 ZS sound card
- Enermax 550 Watt ATX power supply unit
- Dell 2405FPW 24″ widescreen LCD
The latest BIOS version and chipset drivers were installed on the test system, along with the Catalyst 6.1s. Windows XP Professional (SP2) was configured to have Automatic Update, System Restore, and all unnecessary startup services disabled. As a standard baseline, 32-bit color and trilinear filtering were used during testing, and the Mipmap detail option placed at high quality in the control panel; SimHQ does not believe the typical sim-head buys high-end video cards for compromised image quality, though this will result in benchmark scores slightly more conservative in their numbers compared to those seen elsewhere (if driver defaults were left unchanged). The licensed version of Fraps 2.7.2 was used to record benchmark scores for applications that lacked the option to record frame rates, and testing was conducted using the following software:
- 3DMark05 (v120)
- Lock On: Flaming Cliffs (v1.11)
- IL-2: Pacific Fighters (v4.02)
- MS Flight Simulator 2004 (v9.1)
- Splinter Cell: Chaos Theory (v1.05)
- Call of Duty 2 (v1.01)
- rFactor (v1070)
When possible, application anti-aliasing was used rather than forced via the driver control panel. Each simulation/game was initially tested at 1024×768, 1280×1024, 1600×1200, and 1920×1200 without anti-aliasing or anisotropic filtering enabled, and then the X800 XT and X1800 XT were tested again with 4x anti-aliasing and 8x anisotropic filtering as a high quality test category; as a mainstream part, however, the X1600 XT’s high quality settings instead consisted of 2x AA and 4x AF, so please bear this in mind while surveying the benchmark scores. The resolution of 1920×1200 was chosen since, as a 16:10 widescreen resolution, it is the highest display option available on the 2405FPW used with the test system, and allows an examination of how well a current high-end graphics board such as the X1800 XT performs beyond the traditional 1600×1200 threshold.
Benchmark Test Scores
SimHQ included Futuremark’s 3DMark05 for examining the theoretical raster and shader performance characteristics of these graphics boards. The individual fill rate (multi-texturing) and pixel and vertex (complex) shader tests were ran at the program’s default resolution of 1024×768.
As noted above, with its far fewer texture and ROP units the X1600 produces the lowest score in the multi-texturing fill rate test, less than one-third that of the X800 and one-fourth the X1800. With the clock speed similiarities of the new parts, this score tracks fairly well with the number of functional raster units in each graphics chip, particularly with the memory bandwidth advantage the X1800 possesses. In contrast to the first score, however, the X1600 fares well against the X800 in the pixel and complex vertex shader tests, and actually outperforms the X800 in the latter test (largely due to its higher clock speed). The reduced number of pixel shader quads, though, cause it to fall behind the X800 in the pixel shader test by roughly 25%, despite its higher clock speed. That said, the X1600’s scores in these synthetic tests show it performing quite well for a mainstream part; particularly, as stated earlier, with the shift in graphics hardware performance moving away from raw pixel and texture fill rates and toward arithmetic processing. Keep this in mind while examining the X1600’s game scores.
As for the X1800 XT’s scores in these tests, the board outperformed the X800 the most in the pixel shader test. The fill rate and vertex shader scores both showed roughly a 25% advantage for the X1800, while the pixel shader test demonstrated slightly over 66% gain, far exceeding the clock speed variance between the two parts. We’ll see if actual game performance follows this pattern and learn whether or not the X1800’s ultra threaded design allows it to outperform the X800 by more than its 25% clock speed advantage.