How Green is Your PC? Estimating Power Usage Effectiveness

Tuesday, May 12, 2009 – 3:00 AM

Evaluating data center efficiency.How efficient is your PC? Here’s a thought… How about taking some ideas from the peopl e who run data centers? Turns out data centers use a measure called Power Usage Effectiveness (PUE) to assess how energy efficient they are:

Power usage effectiveness (PUE) is a metric used to determine the energy efficiency of a data center. PUE is determined by dividing the amount of power entering a data center by the power used to run the computer infrastructure within it. PUE is therefore expressed as a ratio, with overall efficiency improving as the quotient decreases toward 1.

Err… So what does that mean for your PC? What if you think about the PC on your desk as a mini data center. How would you work out it’s PUE?

Calculating PUE

To calculate a system’s PUE you need to measure the power being drawn from the wall and then estimate all the losses to figure out how much power is actually being used to run the bits of the computer that matter; the CPU, disk drives and graphics card. Dividing one number by the other will give you an estimate of the PUE. That’s the easy bit, the hard part is coming up with the power numbers.

P3 International P4400 Kill A WATT Electricity Load Meter and Monitor Well the first thing to do is measure the amount of power being used by your system. For that you need a P3 P4400 Kill A WATT Electricity Load Meter. It connects between the wall socket and the PC’s plug and measure the power being drawn by the PC.

You probably want to do this when the machine is idle and under full load as the PUE will vary. This is also an excuse to “load” your machine by spending a few minutes killing zombies to get the processor, graphics card and PSU working hard. Alternatively you could use some burn in software like PassMark BurnInTest, that would be a bit dull but probably more accurate. Try and get an average reading from several minutes use rather than just a single figure.

After you’ve done that you should have you two numbers for full load and idle power consumption. Depending on your system these are probably lie somewhere between fifty and a few hundred watts.

A typical load-efficiency graph. Unfortunately not all of the power drawn by your PC makes it to the parts that matter. Power supplies aren’t perfectly efficient. So the next thing to do is factor in the computer’s power supply (PSU) efficiency. You can probably get this from the manufacturers web site. If you’re lucky they’ll give you a graph showing the efficiency under different loads like the one on the right for my Corsair TX650W. If not pretty much all manufacturers will give you an average number, especially if its 80 Plus rated. If not you can take the average number quoted.

What’s interesting about the graph is it shows how it’s possible to use an efficient PSU inefficiently by mismatching it to the rest of the PC. The efficiency peaks and drops off for very low and very high power consumptions.

For example for my my developer/gaming machine the power reading under load is 240W, this is 240/650 = 37% of the PSUs maximum load. From the graph above this gives an efficiency of 83% for a 110V circuit so 240W * 83% = 199W is actually delivered by the PSU to the rest of the computer. Now you know the amount of power actually being delivered by the PSU.

Typical layout of a development or gaming PC. Unfortunately some of the components inside your computer aren’t actually contributing to the bottom line, they’re just there to cool the bits that are. Namely… the fans.

In order to work out the PUE the power consumption of each fan has to be removed from the equation. On the left is a diagram of my developer/gaming machine at home. It has a lot of fans, eight in fact.

I actually looked up the specs for the individual fans, most manufactures give you a range of current requirements and voltage for their fans you can multiply these together to get the power. For variable speed fans you’ll end up with a power consumption range. If you just want a quick answer without looking at a lot of specifications then a reasonable rule of thumb is that most 120mm fans use between 1.5 and 2.5W depending on the speed they are running at.

Now you can work out the PUE. Here’s an example for the machine above:

  Idle Under Load
Power drawn 150W 240W
Power from PSU 150 * 82% = 123W 240 * 83% = 199W
Power used by fans 12W 15W
Power used usefully 111W 184W
PUE 1.35 1.30

How does this compare to other machines or a datacenter?

My Windows Home Server build was specifically designed to be a low power server. What’s it’s PUE? Here’s the same calculation for the WHS:

  Idle Under Load
Power drawn 37W 44W
Power from PSU * 37 * 80% = 29.6W 44 * 80% = 35.2W
Power used by fans + 6W 10W
Power used usefully 23.5W 25W
PUE 1.57 1.76

So the WHS doesn’t use much power but it’s also not that efficient. In fact the lower the overall power consumption of your PC the more the fans are likely to contribute to raising the PUE. The real solution for low power devices is passive cooling. I built the WHS with off the shelf parts so it uses conventional fans. In addition the WHS’s power supply is poorly matched to the system. Most of the time it’s loaded at barely 10%. I’d be better off with a much smaller PSU if I could find one.

In comparison Microsoft’s container based Chicago Data Center has an average PUE of 1.22 and a peak of 1.36. The industry standard is considerably higher than this, around 2.0. But when data centers measure PUE they take the total power usage for the facility and divide it by the power used by the servers. In other words the fans and any power losses inside the server don’t effect the data center PUE. Obviously some big players like Google and Microsoft are trying to drive up the efficiencies of individual servers because they’re running literally millions of them. For the rest of us you’re on your own.

Improving PUE

In the long term manufactures like Dell and HP should simply publish the PUE for their systems. Next time I buy a desktop, laptop or server I should be able to get idle/load power consumption and PUE figures for it. Currently all Dell seem to give me is a rating for the system’s PSU and it’s efficiency, which basically don’t tell the consumer anything useful.

In the meantime, what should you think about when building or buying a new PC and how can you improve the PUE of your PC?

  • Make sure you buy or build a PC with the most efficient PSU you can find. A 1% improvement in PSU efficiency is the same as removing a fan. Visit 80 Plus to find out more.
  • If you’re building a machine make sure that the PSU is matched to the rest of the system. From the power efficiency curve above. Don’t be persuaded to buy an 850W PSU unless your machine really warrants it. A big PSU will be inefficient at low loads.
  • Turn on fan speed control. Most BIOS’s support some form of fan control. Software like SpeedFan can give you more control over how the cooling in your PC operates.
  • Turn on any other power saving features like processor idling – AMD’s Cool’n’Quiet or Intel’s SpeedStep – you need less cooling if you generate less heat.
  • Put your PC into sleep or hibernate mode automatically when it’s not in use. This saves almost as much energy as turning it off (if your PC is plugged in it still draws almost as much power as when in sleep/hibernate mode). Find out more about sleep/hibernate mode here.

Well hopefully that gives you something to think about and some tweaks you can make to your hardware to make is slightly more efficient.

* The WHS build PSU is a Seasonic S12 330W. It has an efficiency of 82% at 20% load. In fact the WHS loads the PSU at around 10% so I assumed the minimum efficiency of 80% based on the power curve for the Corsair.

+ The WHS has five fans – two on the CPU cooler, two in the case and one on the PSU – but these never run at anything more than low speed in normal use and actually shut down for some of the time. These numbers are estimates.

  1. 4 Responses to “How Green is Your PC? Estimating Power Usage Effectiveness”

  2. I can’t see how this number is useful in determining the efficiency of a PC. The losses are arbitrarily separated in ‘useful’ losses such as those inside the conductors of the CPU, and ‘bad’ losses, e.g. in the conductors of the PSU. The proposed efficiency index can then be improved by decreasing ‘bad’ losses, but equally by increasing ‘useful’ losses (such as in the on-board power supply to the CPU). What really matters for the efficiency of the PC though, namely how much work (information processing) the PC gets done, is not factored at all. Indexes such as dividing benchmark results by power draw, as seen on some hardware review sites, arbitrary though they are, come probably much closer of what would commonly be understood as efficiency of a PC.

    I think intuitively you are agreeing, because you advise to turn on SpeedStep/Cool’n’Quiet or suspend/hibernate the PC. These measures actually deteriorate (increase) PUE by decreasing ‘useful’ losses. A hibernating PC will have an extremely bad PUE, as there are almost no ‘useful’ losses at all.

    By JanW on Oct 28, 2009

  1. 3 Trackback(s)

  2. May 17, 2009: How Green is Your PC? Estimating Power Usage Effectiveness | #2782 … | SoftRetails.Com
  3. Jun 8, 2009: The Power of Software : The Data Center in Your Home
  4. Jun 25, 2009: Green Computing isn’t the same as Low Power Computing | #2782 - Thinking about agile (small 'a') software development, patterns and practices for building Microsoft .NET applications.

Sorry, comments for this entry are closed at this time.