(Part 6 in a series of posts on small-form-factor computing)
When we last stopped, we were down to SG08-size, which fits a mini-ITX motherboard and a full-size graphics card. And we were about to remove that graphics card to see how much smaller we could go.
Let’s say we remove that graphics card. What’s left? Mainly just the CPU cooler, and the power supply. We can’t get much smaller without shrinking either of these. And so we’re back to the question we asked in Part 2: Does it need to take up so much space?
The SG08 uses an SFX PSU rated for 600W. Without a graphics card, do we really need that much power? According to various review sites, the 4th-generation i7 processors from Intel have peak system power consumption around the 100W range. I omit power consumption figures for 3rd-generation i7 processors because those are lower compared to the 4th-gen processors.
What’s with the huge difference in numbers? #
Unfortunately, reviewers carry out their measurements with different system configurations. Using a different motherboard can raise or lower power consumption by up to 20W. Even with the same motherboard, changing the performance profile can also affect power consumption significantly.
The way different reviewers load the system varies as well. Anandtech, Techreport, PC Perspective, and x-bit labs use video-encoding applications to simulate more typical CPU loads, while Bit-tech uses a synthetic benchmark for calculating prime numbers, which tends to stress the CPU more. I have picked power consumption figures for real-world benchmarks where applicable.
Although these benchmarks stress mainly the CPU, while keeping storage and graphics at minimal load, these subsystems still use a significant amount of power. Hard disks (HDDs) can use up to 10W under heavy load, while SSDs can use up to 5W at load, but typically less. The high-end graphics cards also use a non-negligible amount of power, as the two numbers from Techreport show.
One last thing to note is that all the above reviewers use system configurations with PSU ratings of at least 400W, typically 600W or more. This is significant, because PSUs have a load range at which they are most efficient. Go out of this range and efficiency plummets, often drastically at very low loads.
PSU efficiency #
Many DIY PC enthusiasts know that PSUs have optimal efficiency around 50% load level. This means that a PSU rated at 800W maximum is most efficient (up to ~90% for high-end PSUs) if the system is drawing around 400W. The optimal efficiency zone tends to fall into the 30–70% range, and if you want to stay above 80% efficiency you will want to keep in the 20–80% range. These are ballpark figures, and vary between manufacturers and PSUs.
If you’re using a traditional desktop and are familiar with issues surrounding PSU efficiency, at this point you will probably stop me to ask why any of this matters. The only significant consequence of lower efficiency is a higher power bill, isn’t it? Let’s remember what we’re talking about here—shrinking the desktop, not ‘green’ concerns.
Without a graphics card, the i7 processors we are looking at will reach no more than ~20% load with a 500W PSU! We are using way overpowered PSUs for a system without a graphics card. This matters when it comes to PC allometry: when we went from 2 graphics cards to 1 graphics card, we were able to go from a larger ATX PSU to a smaller SFX PSU. And likewise here, we are missing out on another opportunity to shrink the PSU.
What are our options when we go below SFX? A decade ago, the only options would be TFX PSUs and 1U server PSUs. Neither are readily available today, and both are still significantly bulky compared to what is possible now.
The 120W threshold #
What is using up all that space in a PSU?
Like with a tower desktop, most of a PSU is empty space, capped by a cooling fan that blows air over the huge heatsinks. Heat is produced because power supplies are not 100% efficient: Even the most efficient ones for desktops don’t go above 95% efficiency. For a 600W PSU, that means that at 50% load (300W), at least 15W is being wasted, i.e. drawn from the socket but not being supplied to the desktop. This wasted power is dissipated as heat, and quite a significant amount it is: in comparison, entertainment tablets typically draw no more than 8W at full load.
How do we tweak the PSU so that it doesn’t need all that stuff for cooling? By reducing waste heat to the point where we don’t need a cooling fan, of course. In addition, when we don’t need to supply so much power, the components we use—capacitors, inductors, heatsinks, etc—can also be smaller. And we’re already familiar with such PSUs:
That’s right, I’m talking about the ubiquitous “power brick” AC adapter.
A desktop PSU converts 110V/220V AC to multiple voltages (12V, 5V, 3.3V DC), in addition to filtering out electrical noise, among other functions. The AC adapter, on the other hand, does mainly one conversion: 110V/220V AC to a single voltage, typically 12V or 19V DC. It’s designed much more compactly, with less circuitry. And you’ll also notice that it is almost impossible to find AC adapters with an output above 200W: beyond that, you start to need huge heatsinks and cooling fans to dissipate the waste heat.
If you’re still having trouble reconciling the idea of an AC adapter powering a desktop, it might help to remember that gaming laptops use about as much power and run on 180W AC adapters.
We’re still missing something to go from 12–19V DC to the voltages required by a desktop (12V, 5V, 3.3V DC). Enter the picoPSU:
The picoPSU takes in DC power from a power brick, and outputs power directly to the desktop motherboard via the connector soldered to the mainboard. The variant shown above is the most common model with peak power of 120W, although it is also available in variants up to 160W; DC-DC adapters up to 200W are also available. But I typically refer to this configuration of AC adapter + picoPSU as the 120W configuration, since it’s the most commonly used one.
We can drastically shrink the PSU from SFX to a power brick + picoPSU, as long as we stay below the 120W threshold.
Cutting it close? #
Whoa, whoa, hold on—the power consumption numbers we just saw were dangerously close to 120W, weren’t they? Is it really safe to run a 4-th generation i7 on just 120W? What if there are fluctuations in power consumption? …
As I mentioned earlier, almost all reviewers take their numbers with a system running on a high-powered PSU. As they take those power consumption figures, their configurations are running below 20% PSU load, often as low as 15% or less (with an 850W PSU)! At those levels, PSU efficiency plummets below the typical >85% efficiency, and can drop to 70% or lower. This means that if we switch to a 120W config (power brick + picoPSU), those power consumption figures we see could be lowered by as much as 20%.
Dissatisfied with review configurations, I decided to try a picoPSU on my 3rd-generation i7(-3770), which is slightly underclocked for reasons related to cooling (more on this in future posts). With a Kill-A-Watt power meter, I get 36W at idle and 90W at load (x264 video encoding). This means that my system only hits 75% PSU load maximum, and even accounting for measurement error and power fluctuations, I’m comfortable enough with those numbers to continue using this 120W configuration.
PC allometry revisited: No graphics card, 120W PSU configuration #
In this post, we cut out the SFX PSU, downsizing it to a 120W picoPSU + an “outsourced” power brick which sits outside the case. In the next post, I’ll get to the remaining space-hogger, that flexible space reserved for CPU cooling.
How much space have we cut so far? I’ll put up a more complete gallery in the next post, but meanwhile, a teaser: