MyriMatch power consumption and efficiency
Now let's apply those same power efficiency measurements to a different test case, the MyriMatch proteomics application. We used the same basic configurations as we did for the tests on the previous page, but with a different application and a shorter total time period. Due to MyriMatch's scaling issues on the Xeon 5355 system, we tested with one, four, and six threads to see which is most efficient. Here are the results.

You may have noticed that the Opteron takes quite a bit longer to finish with just one thread than it did in our performance tests. I believe its slow performance with one thread is caused by a poor interaction between the single MyriMatch thread bouncing from one core to the next and PowerNow's dynamic clock scaling. (PowerNow and DBS were disabled for our regular performance tests.) This problem doesn't occur with four threads, where the Opteron systems' MyriMatch completion times are just what we'd expect.

Measured over the entire time span, the Opteron system is again most efficient by a sizeable margin. Multithreading also comes out ahead again, although not by quite as much, since MyriMatch doesn't tend to scale linearly like POV-Ray.

We've taken the idle power consumption data from the trailing seconds of our time period once more. The results look about like they did in our last set of tests, but you can see how the multithreaded results tend to be slightly lower. That's probably because the processors have had less time to cool after the single-threaded runs, and hotter chips tend to consume more power.

When it comes to the energy used only to complete the task, the competition is very tight—so tight that switching from eight FB-DIMMs to four nudges the Xeon 5160 system ahead of the Opteron 2218. Interestingly enough, the Clovertown system looks to be most efficient with six threads, even though its performance doesn't scale all that well from four to six threads.