Single page Print

POV-Ray power consumption and efficiency
Now that we've had a look at performance in various applications, let's bring power efficiency into the picture. Our Extech 380803 power meter has the ability to log data, so we can capture power use over a span of time. The meter reads power use at the wall socket, so it incorporates power use from the entire system—the CPU, motherboard, memory, graphics solution, hard drives, and anything else plugged into the power supply unit. (We plugged the computer monitor into a separate outlet, though.) We measured how each of our test systems used power during a roughly four-and-a-half minute period, during which time we asked POV-Ray to render our "chess2.pov" scene at 512x384 resolution with antialiasing set to 0.3.

Before testing, we enabled the CPU power management features for Opterons and Xeons—PowerNow! and Demand Based Switching, respectively—via Windows Server's "Server Balanced Processor Power and Performance" power scheme.

In order to keep things as fair as possible, we've tested the Xeons with two different memory configurations. The first config is the one we used in our performance tests, with eight 1GB DIMMs. However, that's a relatively high DIMM count (at least for a low-end server or workstation), and we've heard that each FB-DIMM requires about 5W more power than a regular DDR2 module. If true, that means the relatively high DIMM count could tilt the power-efficiency playing field in the Opteron's favor. Fortunately, we could achieve the same total amount of RAM using 2GB FB-DIMMs, and we just so happened to have four of those on hand from our HP workstation. So, below, you'll see results from our standard Xeons configs with eight DIMMs and from an alternative configuration (in the same system) with 8GB of RAM using only 4 DIMMs.

Without any further interpretation of the data, it's obvious that the Opteron system looks very good. Its minimum power use at idle appears to be tens of Watts lower than any Xeon configuration, and its peak power use while rendering is only about 250W. Meanwhile, render times are comparable to those of the dual Xeon 5160 system. But I'm getting ahead of myself. (Ok, one more quick aside: Notice how single-threaded render times tend to vary from one run to the next, as you can see on the Xeon systems with different DIMM configurations. That's an artifact of the tendency of POV-Ray's render thread to bounce around between different CPU cores, which leads to a bit of variance in single-threaded render times—so I wouldn't blame the DIMM configs for that variance.)

Once we have this data captured over time, we can analyze it in various ways. For instance, one simple way to gauge power efficiency could be to look at energy use over our time span. This method takes into account power use both during the render and during the idle time. We can express the result in terms of Watt-seconds, equivalent to joules.

Looked at this way, the Opteron systems are vastly more efficient than any Xeon system, with the four-DIMM Xeon 5160 taking second place. For all systems, using multiple threads to complete the task sooner proves to be more energy efficient than single threading, in spite of higher peak power draw.

I believe this is a potentially useful way of measuring power efficiency, but it's dependent on performing only one action during a set period of time. We can also break power use down into two components, which may prove more precise. The first of those, of course, is idle power, which is almost always a part of the total power picture. Here's how the systems compare at idle, during the closing seconds of our time period.

The Opterons require much less power at idle than the Xeons, even if we give the Xeons the advantage of using only four FB-DIMMs, which shaves off about 22W of power draw. This much lower baseline power consumption makes the Opterons vastly more power efficient when the system isn't occupied. (I was surprised to see this large of a gap between the Opterons and Xeons, by the way, so I put the HP workstation system on the power meter to see how it fared. Its power consumption at idle and under load was very similar to our Xeon test system, though a bit higher, probably thanks to the workstation's Quadro FX graphics card.)

The next step is to consider the amount of energy used to render the scene. Since the different systems completed the render at different speeds, we'll want to isolate the render period for each system. We can then compute the amount of energy used by each system to render the scene, expressed in Watt-seconds. This method should account for both power use and, to some degree, performance, because shorter render times may lead to less energy consumption.

If you're trying to conserve the energy expended in accomplishing a task, there's much to be said for finishing quickly—as our eight-core Clovertown system's first- and second-place finishes attest. Among the four-core systems, the Opteron 2218 proves more efficient at rendering the scene than the Xeon 5160. Once we drop into the single-threaded results, efficiency goes out the window.