PSU efficiency in SFF systems

I have a little confession to make. I am an utter sucker for high-efficiency power supply units (PSUs). But that may be about to end, for reasons I will explain in short order. First, a little background on power supply efficiency. (Feel free to skip ahead if you’ve heard all this already.)

Power efficiency #

Definition #

Power efficiency, simply stated, is the ratio of the power a PSU provides to the power it consumes. If a PSU pulls 100W from the power socket but only provides 75W of that to the computer system, its efficiency is 75%.

Simple enough. So just buy the PSU with the highest efficiency, and be done with it, right? Not so simple.

For a start, PSU efficiency varies depending on how much load is being put on it. At 100% load, a 500W PSU would be putting out 500W to the system, but pulling 625W from the wall socket. At various loads from 10% to 100%, the PSU efficiency can vary from as low as 40% to as high as 90%. For most PSUs, the efficiency peaks at a load of 40–60%, and drops at the extreme ends close to 10% or 100%. This means that it is no longer sufficient to just look at a single number, e.g. max. efficiency or average efficiency, but to look at the whole efficiency curve—a plot of PSU efficiency vs load.

There is an additional caveat as well—PSU efficiency also varies depending on its operating temperature. More on this later.

Certification #

How do we find this efficiency curve? Unfortunately, such information is not provided on retail boxes, or even in technical specifications for consumer-grade products. Instead, what we have to go on is 80 Plus certification. Power supplies submitted for certification testing are tested under 20%, 50% and 100% loads, and PSU efficiency is measured at those loads, with the surrounding air maintained at 23°C. Depending on the standards it meets at those three loads, manufacturers are then allowed to advertise their PSU as having met those standards.

Great. So just go for the cheapest 80 PLUS Gold/Platinum PSU, and we’re done. Unfortunately again, not so simple. As HWS explains in the link, 80 Plus testing is carried out with a surrounding air temperature of 23°C. In reality, most PSUs will be running at a temperature of more than 40°C, usually more than 50°C under higher loads. And at these higher temperatures, PSUs are also less efficient. This means that while your el-cheapo 80 Plus Gold PSU is supposed to be able to hit 90% efficiency at 50% load, it might not do so In Real Life™.

Significance #

Why does PSU efficiency matter again? If your PSU is drawing 100W but putting out only 75W, the discrepant power is dissipated as heat. On high-performance systems, where system load could be as high as 300W, an efficiency difference of 5% translates to extra power draw as high as 15W.

This may be good news in colder countries, but I live in the tropics, and try to do all I can to minimise system heat output. Though you may not feel the difference in an air-conditioned room (which mine isn’t), you are effectively paying much more for this extra heat: your computer system produces an extra 15W of heat which it releases into the room, and your fan coil unit has to pull an additional 15W or more to move this heat from your room to the atmosphere.

In addition, most PSUs have a cooling fan that pulls in air to cool their components. A PSU that is efficient enough could produce so little heat that it can dispense with this cooling fan, allowing one to build a much quieter system.

Measurement #

How do we go about measuring Real Life™ efficiencies then? By using expensive load and measurement tools, naturally. Fortunately for us enthusiasts, there are a number of hardware reviewers who make it their life’s work to document such things. And there are also PSU manufacturers who make it a point to achieve 80 Plus efficiency standards under real-world conditions.

After much googling, I settled on a Seasonic X560, a well-reputed 80 Plus Gold PSU. Its so efficient that under 20% load, the cooling fan shuts itself off—an alarming sight for most people, who might mistake it for fan failure.

But what is not shown in those reviews is how the PSU behaves under 20% load. This unknown information is rather relevant, because Intel’s 3rd-generation Core quad-core processors have power consumption characteristics in this range. Effectively, that means I have no idea how efficient my X560 is at powering my quad-core desktop. Very worrying for a statistics freak like me.

As I am about to learn, even this expensive unit is less efficient under my current desktop load conditions. Shocking, but true.

The Test #

Specifications #

Processor: Intel i7-3770 3.4GHz (3.9GHz Turbo), stock clock speed
Memory: 2×4GB G.Skill Ripjaws DDR3-1333
SSD: Corsair Force GT 120GB

(Really, that’s all I have in it.)

Power consumption #

I measured power draw at the wall socket with a generic power meter bought off eBay. While I have no means to calibrate it properly, the measurements are close enough to expected figures (extrapolated from i7-3770 reviews) that I’d take these numbers with just a pinch of salt. Besides, it does not invalidate the difference in power consumption between sets of measurements.

Power draw is measured under three conditions: idle at desktop, light processor load under video playback performance (software decoding at 1080p), and heavy processor load under video transcoding (1080p video with x264).

I first measured power load with the system drawing power from the X560, and then with it drawing power from the onboard DC-DC supply in my E-Q6 mini-ITX case. The DC-DC supply is fed via DC jack from a generic 120W AC adapter (bought off eBay).

Seasonic X560
Desktop idle: 33W
Light load: 48W
Heavy load: 94W

LR1005 (onboard DC-DC power supply) + generic 120W AC adapter
Desktop idle: 21W
Light load: 33W
Heavy load: 63W

Implications #

I couldn’t believe my eyes either, but a second and third trial gave similar results (which I won’t report here). I don’t have the kind of equipment needed to characterise PSU efficiency, so I can’t really compare the two PSUs under similar power draw. That looks like a huge difference in efficiency. Considering that the X560 is only under a load of 5–17%, vs. 17–53% on the LR1005, this should not be surprising.

The implications for this are pretty significant for anyone looking to build a basic computer system. If it doesn’t have power-hungry video cards, chances are its load on most ATX PSUs is going to be below 25%.

Unfortunately for SFF enthusiasts, the large majority of desktop power supplies are ATX, and the power rating on these go no lower than 400W. The power rating is important here because it is a rough indication of the load range at which it runs relatively efficiently—remember that 80-Plus–certified PSUs are measured at 20%, 50% and 100% loads. If we look only at such certified PSUs, a 400W PSU can be expected to run at stated efficiencies between 80–400W (20–100% load). At power draws below that … who knows?

More recent ITX computer cases now bundle smaller-form-factor PSUs (i.e. SFX, TFX PSUs) with lower power ratings, typically ranging from 250–500W. For well-reputed brands, this in all likelihood indicates better PSU efficiency at the same power load compared to higher-rated ATX PSUs.

But this still isn’t enough. Such options are still niche, and not always readily available. Furthermore, lower-power processors, such as the non-hyperthreaded i5-3570 and dual-core i3-3220, are going to have even lower power draw. This makes it very difficult for anyone to build a bare-bones, low-power desktop system with an off-the-shelf PSU.

Two notable options remain for the low-power x86 enthusiast: the picoPSU (paired with an appropriate AC adapter) is one. Thin-ITX motherboards with onboard DC-DC converters are the other option. These take in power through a DC jack, much like a laptop, instead of the more typical 24-pin power connector seen on the majority of desktop motherboards.

With Intel gradually bringing down the power consumption of their mainstream processors, it makes sense for PSU manufacturers to have an answer to this trend if they wish to retain a presence in this growing segment of the DIY desktop market. With the newly introduced thin ITX form factor, we might see low-power desktops shrink drastically; slim tower systems and all-in-ones are just the beginning. Even if PSU manufacturers don’t introduce these low-power PSUs as retail products, they will always be needed by OEMs for such systems.

I won’t shoot myself in the foot and suggest that the ATX PSU form factor is on its way out; they will always be around for gaming desktops, high-performance workstations, and perhaps some home servers. In the past two years, we have seen tablets slowly supplanting desktops and laptops for media consumption; a similar trend may be happening with small-form-factor systems. A mid-range desktop, sufficient for consumption-centric needs, can be squeezed into roughly the volume of a large cereal box, and there simply is no way to fit an ATX PSU into that. As desktops evolve in shape and size, PSUs will have to adapt to shifting needs too.