Hard Disk Drive Power Consumption Measurements: X-bit’s Methodology Indepth

Our today’s article discusses in detail X-bit proprietary methodology for HDD power consumption measurements. We will talk about power consumption measuring techniques, there cons and pros, and explain how this type of measurements will be done for all our HDD reviews from now on.

by Oleg Artamonov
12/06/2007 | 04:02 PM

Introduction

Hard disk drive storage capacity and performance are traditionally considered to be the main parameters of a hard disk drive that are worth paying special attention to. Of course, both these parameters, especially the second one, have a lot of different aspects that we can look at, but in most cases, the hardware editors’ attention is given solely to these two things.

 

Such hard disk drive characteristic as power consumption has long remained unattended. You may think that it is pretty insignificant: what can a dozen watts really do for the system, if a contemporary graphics card or CPU consume more by a factor of 10x? However, this assumption is erroneous.

First of all, power consumption discussions have become very acute among hardware manufacturers lately. For example, the new Energy Star 4.0 standard indicates that the hard disk drive should consume 7W at the most in idle mode, which is 14% of the total PC power consumption (taking into account advanced CPU power saving modes, 14% of the total office PC power needs in idle mode may be not that much in the end). There are a lot of factors affecting it: environmental concerns, the lack of energy systems power capacity in industrially developed countries, intention to lower power bills… of course, these savings are tiny within a single computer environment, but if we remember that a single office building these days may contains hundreds and hundreds of machines, the end numbers may be pretty significant.

Secondly, and it is more important for us, hard disk drive power consumption equals its heat dissipation, heat dissipation determines its temperature provided all other conditions are equal, and HDD temperature is directly representative of the mean time before failure. For example, if you check out a pretty well-known study made by Google Called "Failure Trends in Large Disk Drive Population" (PDF file, 242KB), you will see that the probability of the new hard disk drives failure does not depend that much on their temperature any more, while the ones that are about three years old have much higher chances of failing if their temperature exceeds 40ºC.

So, by selecting a more economical hard drive, we can ensure that it will heat up less and hence will work more reliably over a longer period of time. It is especially important for compact MicroATX cases, many of which cannot accommodate additional fans for HDD cooling. However, even in large cases you may face an overheating problem if the system features 3-5 hard drives.

Thirdly, hard disk drives are currently used not only in desktop PCs, but also in notebooks: it will take another considerably while before they switch to SSD (Solid State Drive). And although notebook hard drives are by far not the most power hungry components, we shouldn’t disregard them completely: they do require certain amount of battery life as well.

Fourthly, many consumers and computer users buy 2.5” hard disk drives as portable storage devices inside USB chassis. A lot of chassis like that have no additional power supply, while a single USB connector can provide maximum 500mA current. So, some hard drives requiring a lot of power may face serious stability problems or may not be recognized by the system at all.

It is especially interesting to measure the HDDs power consumption because the HDD manufacturers have been working hard on making their solutions as energy efficient as possible lately. For instance, Hitachi has recently announced the launch of their energy efficient Deskstar P7K500 hard drives for desktop computers that use the same power-saving technologies as notebook drives.

Today we are going to point out a few problems you may come across when trying to measure HDD power consumption and offer solutions for them. From now on we are going to use the methodology described in this article on a regular basis in our HDD reviews and roundups.

Testing Methodology

To measure the hard disk drives power consumption as precisely as possible we put together a simple electronic circuitry that allows us to register random currents varying at high frequency. The main problem is that they normally use an oscilloscope for measurements like that, however, it should receive voltage, but not current. So, it means we need a converter to transform current into voltage.

The latter uses two shunts with 0.05Ohm resistance that are connected to the power cable of the tested hard disk drive. As a result, the voltage on the shunt drops by 0.05V per each Ampere of the HDD current. The operational amplifier (LM324N) multiplies the shunt signal by a little less than 20, so that in the end we get the voltage proportional to the current consumed by the HDD scaled as 0.96V per 1A. Moreover, 0 HDD power consumption corresponds to 1.525V output voltage in our circuitry that is why the obtained signal is translated from volts U into amperes A using the following formula:

I = (0.96*(U-1.525)) А

We use Velleman PCSU-1000 oscilloscope that registers the voltage output by the above described circuitry to measure the exact current varying at high speed. The oscilloscope time base setting or sweep speed is set at 0.5msec/div (digitization frequency equals 250kHz, which is enough to register the signal with up to 125kHz frequency), the sensitivity is set at 0.5V/div. Oscilloscope works in auto trigger mode and the taken oscillograms are then submitted to special software for further processing. This software translates the volts from oscilloscope into amperes using the above provided formula and calculates the average and maximum values. At each stage we take 180 oscillograms to ensure precision (each measurement takes 60 seconds, each second the program requests three oscillograms from the oscilloscope). Each oscillogram is 4000 points long, i.e. the end result is calculated based on 720 thousand measurements of immediate consumption current. You may increase the number of measurements if necessary. Since the oscilloscope we are talking about is a dual-channel one, we can simultaneously measure the HDD power consumption along the +5V and +12V power rails using two I/U converters.

The program processes our measurements and reports the average current along +12V and +5V power rails in amperes (and the corresponding power in watts) as well as maximum registered current values.

The above described circuitry is connected to the hard drive right inside the computer – into the feed circuit. As a result, we can measure the HDD power consumption under any type of workload emulated in such tests as Intel IOMeter, for instance.

Multimeter vs. Oscilloscope

I am sure that some of you will wonder why we need all this complex stuff: amplifier, oscilloscope, additional software? Why all this when we could simply take a regular digital ammeter or multimeter and measure all necessary current with it?

Well, unfortunately, you can obtain more or less adequate results with a multimeter only in idle mode, when the HDD heads do not move. To illustrate this statement we took an oscillogram for Maxtor Atlas 15K II HDD consumption current when working in Intel IOMeter Random Read pattern. Red color stands for the current along +5V power rail, blue color – for +12V power rail, 0 is marked with a black horizontal line and horizontal sweep equals 5msec/div:

The main part of the power the HDD consumes along +12V rail is spent on moving the heads; impulses come in pairs: the first one corresponds to the beginning of the head movement (speeding up) and the second one to the end down of the movement (slowing down). The distance between them varies from almost 0 to the time it takes to move the head from one side of the drive to the other, depending on how “lucky” the HDD is with the sequential requests. Before the heads start moving, we can also see increase in power consumption along the +5V rail indicating that the HDD electronics dealing with the next request activates.

However, we are primarily interested not in the HDD mechanics, but in the impulse characteristics. As you see, first, their amplitude is pretty high (about 4-5 times higher than the constant), second, the front is almost vertical while the entire impulse may be less than a millisecond. How big is the chance that your multimeter will catch this peak?

Unfortunately, there is hardly any chance at all. Multimeters are devices that are intended to work with constant voltage (and constant current, respectively). They simply do not use fast ADC, because it doesn’t make any sense. A typical multimeter measures at a pace of a few tenths of a second, which is 20x higher than what a current impulse from the moving heads is.

To make this explanation more illustrative we translated the oscillogram above into a spectrum:

As you see, there is a big peak at 0 (direct current level), relatively high and more or less steady level in the interval up to a few tens of kHz, high jump to 42.8kHz and another one at 85.6kHz. So, we need a device that can work with at least up to 100kHz frequencies in order to measure adequately the parameters of a signal like that, and multimeter is definitely not the one for the task.

To check out this theory we used two almost randomly selected multimeters – and inexpensive Mastech M890G and a more serious Uni-Trend UT70D. The latter also allows to detect the average, minimum and maximum values within a given period of time.

So, let’s run IOMeter, Random Read pattern, again on Maxtor Atlas 15K II HDD and see what our multimeters will show. Since each of them can only measure one parameter (unlike the dual-channel oscilloscope), we connected them to +12V rail.

It is pretty hard to figure something out on the first one, the Mastech M890G: the values on the screen keep jumping up and down hitting maximum 2.9V and dropping to minimum 2.4V. Using the formula above, we can easily translate these values into consumption current: from 0.84A to 1.32A. This is where we can already see that the multimeter is not giving out the correct values: the oscillogram above shows clearly that the difference between maximum and minimum values is much bigger than 1.5 times. As for the average value, we couldn’t make it from the jumping readings on the screen.

Luckily, we have another multimeter, UT70D that can calculate the average value. Moreover, it can also transmit the reading to your PC via RS-232 interface, so the results will be given in the form of a screenshot:

On the left you can see a window of our own software that processes the oscilloscope data. On the right – the window receiving multimeter readings. The large digits stand for the average value, the maximum and minimum are given in smaller font below. We switched the multimeter to calculating the average value mode at the same time our own proprietary software was launched. It stayed in this mode for 60 seconds that we needed to get all the oscilloscope readings.

So, the multimeter reported the following: 1.06A average consumption, 1.13A maximum consumption. The oscilloscope data reads: 1.04A average consumption and 2.71A maximum consumption. As you can see, the multimeter managed to get the average value pretty closely, but failed to catch any of the consumption peaks.

At the same time, we cannot claim that any digital multimeter will register average consumption correctly: we managed to prove experimentally that our particular UT70D model showed very realistic number on our particular hard drive. We don’t know if the reading from other multimeters or the same multimeter on other hard drives will be adequate, as well.

And of course, it makes absolutely no sense to try detecting peak values with a multimeter. In our experiment they were not even close to the real ones. Moreover, if your multimeter shows high numbers it doesn’t mean they are correct. You will only be able to verify this by comparing the readings with the fully-fledged oscilloscope measurements. But if you have a system like that why use a multimeter at all?