Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
?
EE Times-Asia > Controls/MCUs
?
?
Controls/MCUs??

Advanced software boosts PC efficiency measurements

Posted: 18 Feb 2008 ?? ?Print Version ?Bookmark and Share

Keywords:PC efficiency measurements? 3D modeling?

By Dave Salvator and Jeff Reilly
Intel Corp.

There's new software to simultaneously measure the energy efficiency and performance of AC-powered, PC-based systems, i.e., the Energy Efficient Performance, or EEP. SYSMark 2007 Preview Edition (using EEP v. 2.0), with updated application versions and workloads to SYSMark 2004 (using EEP v. 1.0), now also runs on Windows Vista as well as Windows XP to help you evaluate which machine consumes the least energy as well as which machine provides the most performance for the energy used. Here's a review of the EEP methodology as a whole, the differences between the old and new softwares, and how to get the most from the new version.

(Click to view image.)

Methodology review
The model, from BapCo, an industry consortium comprising companies with a mutual interest in comparable performance evaluation, has two major components: the workload and the usage context. The workload is the usage simulation provided by SYSmark 2007, which includes office productivity, e-learning, 3D modeling and video content creation tasks. It includes a range of activity levels from high-intensity work to idle time (as though a user was answering phone calls or pausing to think). It even assumes human-level typing speeds (Figure 1).

The second component of the model is the usage context, or what we call the "workday." SYSmark 2007 Preview Edition provides a performance measurement. EEP v. 2.0 provides a framework for taking the performance information and combines it with a methodology for obtaining relevant power and energy consumption metrics. This provides the tester with a way to estimate how much energy would be consumed by the equipment in performing a typical day's work.

Note that in both EEP 1.0 and EEP 2.0, we advocate gathering the data from three single separate runs of SYSmark and using the performance and power measurements that come from the median performance run as the basis for the model.

Similar workday, new model
SYSmark 2004 SE was used as the basis for EEP 1.0. With this benchmark, the workday was modeled like Figure 2:

(Click to view image.)

SYSmark 2007 Preview is a different benchmark from SYSmark 2004 SE and has a different organization and total runtime. The change in the model is outlined in Figure 3.

The machine performs an amount of work similar to what it would perform under EEP 1.0, but allows the EEP model to now use both the latest applications and the latest operating system.

If a system takes more than 75mins to run SYSmark 2007 Preview Edition, it cannot be evaluated using EEP 2.0. The model requires each four-hour work period (one in the morning and one in the afternoon) to contain at least three SYSmark 2007 runs and one 15-minute break. Three 75-minute runs plus a 15-minute break equals four hours.

Running the test
One important change to the methodology is necessary when using Windows Vista, Microsoft's newest OS. Windows Vista tunes performance dynamically. This can mean more optimal performance characteristics but can also increase variability among SYSmark runs. Therefore, we recommend performing at least three runs and taking the median when using Windows Vista.

We measure the average power consumed by the system during idle and standby after capturing how long the SYSmark segments take to run and the average power consumed during that time. For idle, we allow the system to rest for seven minutes with no applications active; the system takes power samples over a 3-minute period and calculates average power based on those samples. For standby, we select Standby mode from the Windows shutdown menu; then, the system takes samples over a period of three minutes and averages them.

As in EEP 1.0, there are a few additional considerations to keep in mind:

Use the right tool for the job: We continue to recommend using a wattmeter capable of continuous data logging. The meter must be set at an adequate sample rate; we used a rate of two samples per second (2Hz) for the calculations presented here.

Be aware of variability: Even supposedly identical integrated circuits and PC components can vary a great deal from unit to unit. Vary only one element at a time and designate a "unit under test." If testing pin-compatible CPUs, for example, do so using a single computer, installing and removing each CPU in turn.

Use external monitors when testing notebooks: When using EEP 2.0 to calculate the power draw of a notebook on AC power, connect an external monitor and set the notebook's graphics controller to use only that monitor. This removes the energy used by the notebook screen backlight, typically the largest energy consumer in such systems.

Keep metrics separate: Performance metrics and energy cost metrics should not be combined, as linear correlation between the two cannot be assured.

Enable power management: Generally, any energy-saving technology available on the system should be enabled as it would in a true end-user scenario.

Don't round prematurely: Kilowatt-hour figures collected during the test runs will be small decimal numbers that need to be multiplied by larger numbers. Resist the temptation to round off until after the final energy cost calculation has been made.

Use locally appropriate energy cost: Energy cost varies by country. Make sure to substitute an appropriate value for your location. In this example, we use 7.705 cents/kWh (the average of industrial and commercial energy prices in the U.S. as of May, 2007, according to the U.S. Dept. of Energy).

Don't log during reboots: The wattmeter should be turned off when the system reboots between each of SYSmark 2007 Preview Edition's four test components. Measuring energy consumed during rebooting will skew the results, as users do not typically reboot their systems several times a day.

Doing the numbers
EEP 2.0 required capturing the following data for each run:

  • The SYSmark 2007 Preview overall score

  • Average system power-draw during the Video Content Creation test run

  • Average system power-draw during the Office Productivity test run

  • Average system power-draw during the eLearning test run

  • Average system power-draw during the 3D Modeling test run

  • Completion time of the overall test run (sum of the four sub-test runs)

  • Idle system power-draw

  • Standby system power-draw

In this example, we measured the performance and energy cost for the Intel Core 2 Duo E6700 processor with integrated graphics on the Intel Desktop Board DQ965GF motherboard. Both Enhanced Intel SpeedStep Technology and Minimal Power Management are enabled. Also, note that measurement best practices would dictate taking the median of three runs, and testing under Windows Vista would suggest the use of the median of five runs.

To convert to kWh, we divide by 60, to make the units W-hrs, and then by 1,000 (or, in a single operation, divide by 60,000): 3,028.104 watt-minutes/60,000 = 0.0504684kWh.

Now, we calculate energy use during the morning period. First, we determine the amount of time remaining at the end of the morning period. Recall that the first 30mins of this time the system will be idle, and then it goes to sleep:

Total: 135.9mins.

That leaves 104.1mins remaining in the morning work period. The first 30mins of that will be at idle, so there will be 74.1mins of sleep during the morning work period. Now, we figure out the energy usage during the idle periods (15min break and 30min idle) and the sleep time of 74.1mins.

The total Idle time is 45mins, and the system idles at 59.6W. So 59.6W x 45mins = 2,682W-mins. Converting to kWh, we get: 2,682W-mins/60,000 = 0.0447kWh.

Our example draws 3.2W when asleep. Now, we figure out the energy consumed during the 74.1 minutes of sleep:

3.2W x 74.1mins = 237.12W-mins. Converting to kWh, we get: 237.12W-mins/60,000 = 0.003952kWh. So, summing all the components of the morning work period, we have:

Next, we calculate the energy used while the system is asleep during the lunch hour: 60mins x 3.2W = 192W-mins. That equals 0.0032kWh.

The afternoon work session is identical to the morning session, so the complete workday consists of:

Finally, we calculate the energy used overnight while the system is sleeping. There are 15hrs between 5 p.m. and 8 a.m., so: 900mins x 3.2W = 2,880W-mins, or 0.048kWh. The total for the 24hr period of a workday is thus:

Next, we calculate the energy used during non-workdays, such as weekends, holidays and vacation. The system is assumed to be asleep on these days, and there are 1,440mins in 24hrs, so: 1,440mins x 3.2W = 4,608 watt-minutes, or 0.0768kWh.

In the EEP 2.0 model, we assume there are 240 workdays, and 125 non-workdays, so the annual energy consumed is: (Workday energy used x 240) + (Non-Workday energy used x 125) = total annual energy. That's numerically equal to (0.4513144 x 240) + (0.0768 x 125) = 117.915456kWh.

The annual energy cost for our example system is therefore 117.915456kWh x $0.0705 = $9.09

The following figure shows costs calculated for various other Intel processors in the same unit under test using EEP 2.0.

As with EEP 1.0, EEP 2.0 is just a starting point for discussion. Measuring energy consumption relative to performance is a task of interest to a large number of industry players as well to consumers of computer equipment. Until an industry-wide standard is agreed upon, we will continue to update the EEP model.

Tested System Configuration Details: Intel Core 2 Duo E6700, Intel DQ965GF motherboard with Intel G965 Express chipset; graphics driver version: 7.14.10.1244; chipset driver version: 8.3.0.1013, 2x1GB Micron DDR2-667 5-5-5-15, Seagate Barracuda ST 3320620AS 320Gbyte/16Mbyte cache/7200rpm, Windows Vista Enterprise, DirectX 10. Tests conducted with Enhanced Intel SpeedStep Technology power saving feature enabled.

Click here for more information on energy-efficient performance. Intel has an online calculator based on the methodology proposed here. Click here for further discussion of the proposed methodology.

About the authors
Joining Intel in 2005, Dave Salvator is the worldwide client capability evangelist with the performance, benchmarking and analysis group at Intel Corp. He is responsible for developing and promoting the Intel Digital Home Capabilities Assessment Tool, and runs the Intel Capabilities Forum Web site (www.intelcapabiliitesforum.net). In addition, he has done extensive work in the area of energy-efficient performance, which measures platforms both in terms of delivered performance and associated energy cost.

Jeff Reilly is a principal engineer in the performance, benchmarking and analysis group at Intel, where he is responsible for measurements and analysis of Intel's client products. Since joining Intel in 1990, he has also worked on workload and benchmark development, having been involved in many of the industry consortiums that develop benchmarks, including the TPC, BAPCo and GPC. Currently, he is actively involved with SPEC on their CPU benchmarks suites, having chaired the SPEC CPU subcommittee since the early 1990s.




Article Comments - Advanced software boosts PC efficien...
Comments:??
*? You can enter [0] more charecters.
*Verify code:
?
?
Webinars

Seminars

Visit Asia Webinars to learn about the latest in technology and get practical design tips.

?
?
Back to Top