MPG for CPUs
While these differences may be striking, the watts per hour measurement in and of itself isn't all that meaningful. I could tell you that my car consumed two gallons of gas in an hour, to which you might want to know how far I actually drove in that time frame. In other words, how many miles did I get to the gallon? (You might also wonder about other factors that might affect MPG, such as whether I was driving a hybrid or an SUV, what the driving conditions were, and whether I was hauling my dry cleaning or a ton of bricks.)
In his test, Nelson calculated the equivalent to MPG for servers by dividing their throughput by how much power they consumed to yield "transactions processed per watt hour," or TWH; in other words, Nelson calculated their power efficiency. Here, Nelson found that the Opteron-based server delivered between 12.6 and 26.8 percent better power efficiency than the Xeon-based machine. (I should once again note that the Xeon clock speed was also slightly higher than that of the Opteron, which could be a factor here.)
For example, at 500 simulated users and with 4GB of memory, the Opteron machine had a TWH rate of 1,686.2; the Xeon machine's TWH rate was 1,382.9. The difference here was 21.9 percent. At 16GB of memory and 500 simulated users, the Opteron's TWH rate jumped to 1,831.2; the Xeon's was 1,433.7. The difference here was 27.7 percent.
There's one other notable finding in Nelson's test: The Opteron-based server consumed substantially less power than the Xeon-based server when the systems were idle. At 4GB of memory, the Opteron consumed 164 watts per hour; the Xeon consumed 198.8 watts per hour. That's a 17.5 percent difference. At 16GB of memory, the Opteron consumed 169.6 watts per hour at idle whereas the Xeon machine consumed 222.8 watts per hour, a difference of 23.9 percent.
Why's this important? As Nelson puts it, "Most servers spend the vast majority of their time in a powered-up, but idle, state. File servers, Web servers, and e-mail servers are normally left powered on 24/7, even though most offices are closed and empty 75 percent of every week." If a company must waste money powering and cooling servers that aren't in use, better to reduce that expense as much as possible.
[ Learn how companies such as Cisco and Cassatt are cutting their energy bills by powering down servers when they're not in use. ]