Benchmarks, Settings and Hardware
The benchmarks we used to evaluate performance were games from 2006. When it comes to benchmarking with games, there are two schools of thought. The first is to run with minimal graphical settings to minimize GPU involvement and focus on the CPU, and the second is to operate at high graphical settings to emulate “real life” situations that focus more on the GPU. Both of these rationales have merit so tests were run under both conditions.
In general, the “low” settings used a resolution of 1024×768, low texture/shadow quality, minimal anti-aliasing, and minimal anisotropic filtering. The “high” settings use a 1600×1200 resolution, maximum texture/shadow quality, maximum anti-aliasing, and maximum anisotropic filtering (see Chart 1 below).
Chart 1 – Benchmark Settings
Originally this review and analysis used four different gaming benchmarks: Far Cry, Prey, FEAR and X3: Reunion. Unfortunately, the results from X3: Reunion with CodeAnalyst were so problematic as to be worse than useless – they were outright incorrect and impossible in some cases. While these have been included in our later graphs for the sake of completeness, the reader is warned that the data on the AMD K8 running X3 is almost certainly wrong and later analysis will not attempt to even postulate the problems with X3. Consequently, we include no information on the configuration of X3 above – however, it was run similarly to the other benchmarks. At the high settings, it used 1600×1200 with maximum effects and on the low setting, it used 1024×768 with minimal visual effects. The bottom line is that we have two meaningful results for each of three specific game benchmarks, one at ‘high’ quality and the other at ‘low quality’.
Thanks to our sponsors at nVidia, Asus, Crucial, and Western Digital, PC Power, Intel and AMD – our testing environments were as identical as possible so that our comparisons are fair. Our two test systems were configured as below:
Chart 2 – System Settings
Neither of these two systems are particularly modern – they were originally set up in late 2006 and kept intact since then. Intel has since upgraded the Core 2 Duo at 45nm to use a newer, refined microarchitecture (Penryn), although it is largely similar to the Core 2 Duo used above. AMD does not produce the 90nm K8 anymore, but it is very similar to the 65nm K8, which still sells in quantity.
Of course, AMD’s most advanced desktop offerings are based on Barcelona, which will be supplanted by Shanghai later this year. Intel is also on the cusp of releasing a new high-end desktop chip, Bloomfield, which is based on the Nehalem microarchitecture. Naturally, these new microarchitectures will present an opportunity for even more analysis and comparisons to the prior generation.
Since this review and analysis explores the microarchitecture of each CPU, a brief review is in order. Figure 1 below shows the microarchitecture of the Core 2 Duo and the K8.
Figure 1 – Microarchitectures for the Core 2 and K8
These diagrams are in no way complete representations of each microarchitecture (for instance, the different branch prediction arrays are not shown). For those who require a more thorough refresher, we recommend our articles on the Core microarchitecture and Barcelona.