AMD's Athlon 64 Has Arrived: the Athlon 64 FX and Athlon 64 (and Intel's P4 Extreme)

Sysmark 2002: Why Use Obsolete Applications?

The well-known Sysmark 2002 from Bapco is very popular, although the mixture of software is hardly up-to-date: MS Office 2002 without service packs, Windows Media Encoder 7.1 - the current one is Version 9 - the table lists them all.

How about service packs for the Sysmark? If Bapco does not release any updates in the next few weeks, we will consider removing this benchmark suite from the test run without delay.

Benchmark Tests - Done Right!

A few comments about our benchmarks at this point. High frame rates such as can be reached with 489.5 FPS (Athlon FX) in the case of Quake III, with 73.07 FPS in the case of Comanche 4 and with 91.1 FPS in the case of Splintercell are the order of the day. However, the lowest graphics resolution is selected, in order to shift the workload more towards the CPU and therefore away from the graphics card. In practice, of course, hardly anyone would play with such settings. But the purpose of our investigation is to show the scaling of a CPU with the respective program.

Since in practice (= real play, user vs. PC) the load on the graphics card is considerably greater, the frame rate with higher resolutions like 1024 x 768 or 1280 x 1024 dots drops more drastically than our benchmark results (obtained at the lowest resolution) would suggest. Moreover, benchmarks - if they are to be replicated 100% - always show a fixed animation process. Factors such as opponents' performance-eating AI (artificial intelligence) naturally cannot be tested because an already played scene is merely replayed. During real play, the AI reacts dynamically to the respective playing situation. The calculations necessary for this would naturally require additional system resources.

Critics of game benchmarks point out quite rightly that it is not just maximum frame rates that are decisive, but the average as well. Who hasn't experienced what happens when a real-time 3-D strategy game runs with a network connection and starts to tremble when large objects suddenly appear, since factors like AI, among others, are ultimately responsible for this action.

For the first time, a real-time strategy game is included - Warcraft III's "The Frozen Throne" (Version 1.12). This test is very cumbersome, as there is no official benchmark. A tournament in 8x speed is replayed on camera, and during a real time lapse of 13 minutes the average image repeat rate is noted with the "Fraps" tool.

The applications we are using as benchmarks show a considerably higher practical experience. Here we are concentrating primarily on high-end applications.