Measuring server power savings before you buy

Why most power-saving trials prove nothing

A server's power draw moves with load, temperature, fan speed and firmware state. Plug a device in, watch the meter for ten minutes, and see a lower number: that is not a result. The load may have dropped. The room may have cooled. A background job may have finished.

A fair test removes those variables. It takes an afternoon of setup and a few days of logging. The method below works for any power-saving product, including the PowerStic and Exodus USB devices that Current RF makes.

Choose where to measure

Measure at the wall, not inside the server. The server's own power reading comes from internal sensors with their own error. A metered rack PDU, or an inline power meter on one server's cord, gives you a number from a source outside the device under test.

Use the same meter for every run. Log real power in watts and also energy in kilowatt-hours over the full run. Real power matters more than apparent power, since the power factor of a supply can change between configurations.

Build the baseline

• Pick one server, or two identical servers if you can spare them, with the same firmware, BIOS settings and operating system image.
• Fix the workload. Use a repeatable benchmark or a replayed job mix, not live traffic.
• Run the baseline for at least 24 hours, so a full daily cycle of ambient temperature is covered.
• Log inlet air temperature next to power. Log fan speed if the platform exposes it.
• Write down the exact steps, so the second run is the same as the first.

If you have two identical servers, run one with the device and one without, at the same time, in the same rack row. That removes drift in room conditions. Swap the roles in a second run to catch any difference between the two machines.

Install the device. Keep the workload, duration, firmware and placement the same. Do not reboot into a different configuration in between unless the install requires it, and if it does, note that in the log.

Run for the same length of time as the baseline. Repeat the pair at least three times. Then compare the difference between baseline and device runs to the difference among the baselines themselves. If the baselines vary by 2% and the device shows 1.5%, you have not found anything.

What Current RF says about its devices

Current RF describes the PowerStic and Exodus as USB plug-in devices that recycle waste energy in computers and network devices. The PowerStic is offered in USB-A and USB-C, and the shop price is $59.99. Exodus has no shop listing and no published price.

On the payback side, Current RF estimates that a PowerStic pays back within three months of operation in a data center. Elsewhere on its cyber-security page the company gives an average return of 3 to 10 months. Both are company estimates. They depend on your electricity rate, your cooling overhead and your device count, none of which Current RF can know in advance. Treat them as claims to confirm with your own meter, not as planning inputs.

The company also publishes application notes on server power reduction. Their titles refer to workstation and server power, total cost of ownership, and Supermicro and HP platforms. Read the test conditions in each one: load level, duration, measurement point, and ambient temperature.

How the loaner program fits

Current RF runs a loaner program for the PowerStic and Exodus. A team applies through the contact form, then measures the saving before buying. That matches the method above: you get the device, you run your own baseline and your own comparison, and you decide on your own numbers.

The company also says it offers on-site power and security testing for data-center and business systems. If your team prefers an outside party to set up the measurement, ask what the test plan covers and who keeps the raw logs.

Turn the result into a decision

Convert the measured watts saved per server into kilowatt-hours per year, then multiply by your own electricity rate and your cooling overhead. Divide the device cost by the annual saving to get payback in months. Use your own figures at every step. If the measured saving is inside the baseline spread, the payback is not defined yet, and the honest answer is that the device did nothing measurable on your hardware.

Details of the data-center application and the loaner request form are on the Cloud Computing page and the Contact page.