Tuning Cool'n'Quiet: Maximize Power And Performance, Part 2

Benchmark Results: Cinebench R10

Unlike the other benchmarks in this article, Cinebench R10 is unique because it stresses the system and processor with different scenarios. There's the OpenGL viewport test, which measures graphics card performance, the single-threaded render test, and the multi-threaded render test, which measures processor performance.

That first part of the graph seems familiar doesn't it? As with SuperPi, performance in Cinebench's OpenGL and single-threaded test suffers when you enable Cool'n'Quiet on the Athlon X2 and Athlon II. In the multi-threaded render test, all cores are utilized--no more thread “jumps.”

Let's look at the test results. Since we’re more interested in processor performance, we're going to focus on the single- and multi-threaded render test results.

We noted before how asynchronous clock changes were adversely affecting performance in single-threaded applications. Cinebench's single-threaded rendering test confirms this. As in SuperPi, “fixing” this behavior returns performance to where it should be--very close to levels without power management. Admittedly, it's still lower, but the difference is small enough not to be noticeable in real usage situations (1–3%); they may even be normal variations between test runs.

The lack of L3 cache and asynchronous clock changes seems to adversely affect performance more on the Athlon II X2 250. It’s more pronounced than on the Athlon X2s. But once that’s not a factor, everything falls into place, and the Athlon II X2 250 takes the lead among dual-core processors due to its higher clock rate. Not surprisingly, the Phenom II X4 955’s 3.4 GHz clock also allows it to take the top position.

The multi-threaded rendering test is basically similar to what we’ve seen before in 3DMark Vantage.

It might be surprising to some, but the results we see here are what we expected when we saw the earlier SuperPi and WinRAR results. Yes, average consumption-wise, the greater number of cores a processor has (and the higher its clock rate), the higher its power consumption. This is true for both single- (SuperPi) and multi-threaded (WinRAR) applications. But both test results also show that if you can get operating voltage low enough, the power consumption differences shrink significantly between a quad-, triple-, and dual-core processor.

With SuperPI, which is purely single-threaded, total power consumption still favors a processor with the least number of cores. In WinRAR, it’s the reverse. These Cinebench results confirms this, and add a new twist. Even with long periods of single-threaded work (on the OpenGL and single-threaded rendering test), once you run multiple threads (and quite possibly, multiple applications), the quad-core processor actually consume less power. There is catch: you need to reduce the operating voltages.

Create a new thread in the UK Article comments forum about this subject
This thread is closed for comments
Comment from the forums
    Your comment
  • Anonymous
    crystalcpuid is a better alternative to AMD's CNQ. Multiplier management allows a wide range of multplier/voltages to be selected. A bit of testing and you can use really low voltages at only slightly lowered multiplier settings. Power consumption is related to the square of the voltage so low voltage give massive power savings. AMD CNQ uses very conservative CPU voltages. You dont need to stick to that ridiculously low idle frequency either and its latency penalties.
  • pertshire
    Great article. Really open my eyes that quads can actually be more energy efficient that dual cores (On application that ends of course)