2. Benchmark Setup

Here's a comparison table of all the systems:

System
Processor	Intel Pentium 4M 2 GHz
Memory	512 MB DDR-SDRAM
Chipset	Intel 845M
Display	1600x1200 TFT
Battery	66 Whr
Graphics Subsystem
	ATi Mobility Radeon 9000	ATi Mobility Radeon 7500	NVIDIA GeForce4 Go440	NVIDIA GeForce4 Go460
Core Clock	240 MHz	280 MHz	220 MHz	275 MHz
Memory Clock	420 MHz (DDR)	400 MHz (DDR)	440 MHz (DDR)	580 MHz (DDR)
Graphics Memory	64 MB	64 MB	64 MB	64 MB
Driver Version	7.75	6.13.10.6052	36.80	36.80
Theoretical Fill Rate	960 Mpixel/s	560 Mpixel/s	440 Mpixel/s	550 Mpixel/s
Peak Memory Bandwidth	6720 MB/s	6400 MB/s	7040 MB/s	9280 MB/s

You can see that while ATi's chips are ahead in theoretical fill rate, they lag behind NVIDIA's products when it comes to memory bandwidth. 3D performance depends much more on memory bandwidth than on theoretical fill rate numbers. That is why the two NVIDIA chips can easily keep up with their ATi counterparts when it comes to the actual frame rates.

Direct3D Benchmarks Mad Onion's 3DMark2001SE has become a de-facto standard these days. We use it for the overall result as well as the fill rate number. AquaNox : We used the retail version of AquaNox and patch 1.17. Comanche 4 Demo Benchmark : We ran the new version 1.0.1.18 Dungeon Siege : The official benchmark patch was used and the file 'system_detail.gas' was adjusted to add the configuration for Mobility Radeon 9000. Max Payne : We ran the demo 'Final Scene 1' for the benchmark.
OpenGL Benchmarks Quake 3 Arena is another de-facto standard, even though it's very old by now. We used good old 'demo001' and version 1.17. Jedi Knight II : We used the retail version and the 'jk2ffa' demo. Serious Sam : We ran the retail version and the 'Valley of the Jaguar' demo.
Anti-Aliasing or Anisotropic Filtering Benchmarks
I decided against benchmark runs with anti-aliasing or anisotropic filtering enabled, because the scores we got with both features disabled were already so low that it would have been unrealistic to think that anyone would actually play games on his notebook with those features enabled.

Benchmarking Screen Resolutions
Most notebooks come with screen resolutions of either 1024x768 or 1600x1200. It's always preferable to run 3D Games in the original resolution of the display panel, as it avoids artifacts due to interpolation. We therefore tested at 1024x768 and 1600x1200 only. In both cases we used 32-bit color.

3D Benchmark Results

3DMark2001SE

Of all the benchmarks, 3DMark2001SE is the one where Mobility Radeon 9000 is furthest ahead of the competition. The reason why is due to MR9000's support of DirectX 8.1. It allows it to run the 'Nature' subtest of 3DMark2001SE, which is something that all the other cards are unable to do. Still, those numbers don't really reflect gaming performance.

At 1600x1200, NVIDIA's upcoming GeForce4 Go460 is reaching the same score as ATi's new mobile 3D chip, even though GeForce4 Go460 is unable to run the DirectX 8 tests of 3DMark2001SE. This hints in the direction that the performance of Mobility Radeon 9000 might be lagging behind NVIDIA's upcoming chip at this high resolution.

A look at the fill rate scores of the four contestants is somewhat sobering. While Mobility Radeon 9000 may still reach the highest score in single texturing, it's only third best in the multi-texturing test. You can see that GeForce4 Go460 looks quite strong here, too.