Does Saving Power Mean Hurting Performance?

Conclusion

In most of the conventional benchmarks that are based on games, applications, or synthetic tests running as fast as possible in the shortest time, we did not find considerable performance differences between the high-performance, balanced-power, and power-saving modes. Most of the benchmarks differ by a maximum of 2%, which translated into differences of just a few seconds for processing workloads and very small frames per second differences for 3D game benchmarks. The reason for this tiny difference is obvious: once the benchmark kicks in, the processor remains in C0 state and hence in active mode for almost the entire time. Just a few benchmarks that require more memory activity instead of permanent CPU peak load (WinRAR) really benefit from switching off power-savings mode.

Where Performance Differs

However, we also ran SYSmark 2007 Preview to analyze the performance differences in workloads that match normal office and consumer usage models and looked at the system peak and idle power consumption. SYSmark is a benchmark geared for popular applications that gauges system performance. Since SYSmark also measures performance when the CPU is in idle mode, AMD and Intel processors will both frequently switch into more efficient P-states using C'n'Q/Enhanced SpeedStep, hence reducing processing performance. The differences SYSmark measured are hardly noticeable during everyday computer use, but they do show very clearly that there is a performance impact at application-like loads.

Power Savings Versus Performance

Since demanding applications don’t show much performance difference and application scenarios are oftentimes more limited by the user than they are by the available resources, we recommend switching on the key power-saving features such as SpeedStep and switching all other items into a balanced mode. This way, you’ll get maximum power savings without losing a lot of performance. Without a doubt, the performance impact will not be noticeable for the vast majority of users.

For power consumption, the system idle power went from 99 to 96 W for the Core 2 Extreme QX9650 and from 92 W to 86 W for the Core 2 Duo. Knowing that all other components remained unchanged, these differences are significant. The differences in peak power were similar, which were 175 W vs. 168 W for the quad-core Extreme processor and 135 W vs. 129 W for the Core 2 dual core.

Performance-hungry enthusiasts and power users might want to switch off most of the power saving items. Everyone else, however, is better off turning these options on to a balanced setting.