How is the power consumption of laptops so drastically different from desktops built with laptop components? I've kept an eye out for minimal power consumption desktops for several years, and seen very few that go below 40W at idle. My personal machine (see Small & Flexible: Back-of-LCD variable power consumption) bottoms out at 37W - with the processor at 300MHz, a laptop HDD, and integrated video. Machines in SPCR's AMD Turion 64 on the Desktop article bottomed out at 36W, and those in the Desktop CPU Power Survey, April 2006 only went down to 40W.

Meanwhile, I've measured 1.6GHz Centrino laptops idling at 8W. Obviously laptops are designed from the ground up to be power efficient, but what are the concrete differences? Since there are folks using the same processors and HDDs, I'm inclined to think it must be related to the motherboard and/or power supply. It seems like it isn't just conversion efficiencies of the PSU and voltage regulator module (VRM), though - if 8W laptops have 90% overall efficiency at idle (a stretch), that means that 37W desktops with similar components would be only 20% efficient overall at idle. I realize ultra-low-power desktops run on the bottom end of the efficiency curve, but the lowest PSU efficiency SPCR has reported at 40W DC (that I've found) is around 65%.

A New Energy Star... in 2007 has some good info on VRM efficiency. At such low power levels, however, I have to believe the CPU is a small part of total system power and as such VRM efficiency isn't going to matter more than 5W.

So what else could be the culprit - mini-PCI vs. regular PCI, specialty mobile video vs. standard integrated video, integrated components (ethernet, USB, IDE) that power down when not in use? Any further info on this would be greatly appreciated!

You've already singled out the power supply. The simpler single-voltage power supplies used by laptops makes it much easier to design an efficient power supply. Such power supplies can also be tweaked for efficiency at the low end since they are typically rated for ~70W instead of 500W.

As for the rest, I believe that much of the savings come from there being far fewer features on motherboard chipsets, and from disabling features when they're not needed.

On a desktop board, you often have on-board sound, RAID, 4+ disc controllers, a PCI and/or PCIe bus, a whole host of legacy controllers, etc, all of which consume power even when you're not using them. Laptops, on the other hand, either don't have most of these or will disable the parts that aren't in use. Laptops don't have RAID, support a maximum of two discs (plus an optical drive), will disable the sound card except when it is actually being used. A lower-powered northbridge can be used, since it doesn't need to able to handle the worst-case scenario of driving a dozen or so PCI devices. They don't have support for serial/parallel ports. I'm sure I'm missing a few as well.

Anyone else know of anything I'm missed?

I don't think they really are that drastically different, if you could really build a desktop with 100% laptop components. For example a Pentium M, Core Duo or Conroe/Merom (when it comes) on an integrated graphics desktop motherboard (which doesn't exist for those CPU's AFAIAA) with a low power consumption notebook drive with a Seasonic PSU should idle in the sub-20W range.

Like Devon said, no doubt notebooks are optimised for power efficiency up to their eyeballs, with umpteen power saving modes, efficient VRM's, various sleep states, high efficiency backlights for the LCD screen, etc.

jaganath - Maybe "drastically" wasn't quite the right word. You're right, it isn't really possible to build a desktop with 100% laptop components. However, I don't think the Pentium M, Core Duo, or Conroe/Merom are the answer. I doubt their power consumption at idle is significantly lower than heavily underclocked/undervolted Athlon XPs, Turion 64s, or Athlon 64s - it's at full load that they really shine. I think the motherboards (integrated graphics or not) and PSUs are the biggest power culprit once you go below ~40W, and no processor/HDD combination will get you under 20W without going to a laptop brick and DC-DC PSU. Even then, the motherboard will probably need more power for all its peripherals than an idling CPU.

I like the way Devonavar put it - since laptops rarely need over ~70W, their PSUs are designed for efficiency in a much lower range than desktop PSUs. And their motherboards are designed with efficiency in mind, with a more efficient VRM and "smart" on-board peripherals that turn off when not in use. Unfortunately, we probably won't see these features on desktop motherboards for a long time, if ever.

Just to drive the point home about low power PSUs, I should point out that the power consumption of most consumer power supplies with no load applied is typically between 10~20W. This is something I've been looking at recently on the test bench. Even the newest revision of the S12 draws 14W with no load applied. With a simpler PSU that only supplies a single voltage (typically ~19V), this minimum power is probably much, much lower.

Devonavar - Great info, thanks. I never thought to test a PSU while on but un-loaded.

A little more digging revealed lowpowercomputing's My Geode PC thread, in which a 120W DC-DC setup with an external brick lowered his overall draw from 44W to 37W (note that his DC-DC PSU died within minutes, however). And that's with a desktop CD drive and non-integrated (ATI 9100) graphics...

Also, josti tested the efficiency of a 110W brick that outputs 17.2V in The efficiency of a fanless 110w PSU kit. Pretty good efficiency above 20W DC output (65%+), but it drops to 51% at 10W.

Assuming lowpowercomputing's PSU was similar, his DC draw is 24W, and his 350W AC-DC PSU is only 55% efficient at that power level.

If my PSU is 55% efficient at my lowest 37W AC reading, then I'm drawing 20W DC. Switching to a brick like the one josti tested would get me an idle draw in the 31-32W AC range. Since all parts in this hypothetical PC are very similar to the power draws of laptop components, I'm thinking the ~23W difference between the Centrino laptop and this hypothetical PC is mostly accounted for by the motherboard and its always-on features.

I noticed that SPCR's new review of the 17" Apple iMac measured only 33W consumed at idle, and that includes an LCD (albeit with the brightness at 0%). Granted, Windows XP was 37W, but that still probably includes at least a couple of watts of standby power for the LCD. Web searches indicate a PSU with a 180W rating, which seems awfully high given the max power SPCR recorded of 73W. Still, lowering the "ceiling" to 180W may have increased the efficiency a bit. Also, due to its integrated nature, the motherboard probably doesn't include IDE, PCI or PCIe buses, etc. - just integrated sound, DVI video, gigabit ethernet, usb, firewire, and one SATA controller for the internal HDD. Regardless, it's good to see that idle wattages are actually going down (the iMac G5 idled at about 45W with the screen off according to this page - Saving Power on idle PCs).

Did the iMac PSU you saw output multiple voltages? I would expect it to be more efficient if it only had to deal with one.

One other thing: The iMac does have an internal PCIe bus for the X1600. Chances are that all of the integrated peripherals also operate off of a PCI bus. I think the difference is one of scale (since fewer devices have to be driven), not one of kind.

I didn't actually see the iMac PSU - just did some web searches to figure out its capacity. SPCR's review (which you had a part in?) says the input power cable is AC, leading me to believe there's no brick.

I don't think the multiple voltages is really that much of a problem - most DC-DC PSUs I've seen (generally 12V DC in, outputting 5V, 12V and whatever the motherboard needs) have efficiencies in the 85-95% range. I would actually think that you would have more options in building an ATX AC-DC PSU since more of it is integrated, rather than having to have two separate pieces (brick and DC-DC PSU).

I still buy your original explanation that a lower maximum power allows designs with better efficiency at idle wattages.

If I'm understanding you correctly, you're saying that iMacs use less power partly because there are fewer peripherals to connect (no RAID, for example), and partly because those fewer peripherals require less out of the chipset, allowing lower-power chips. Makes sense to me anyway...

Question on the video - are all integrated graphics motherboards essentially running a "hidden" connection to the PCI/PCIe bus for the GPU?

Addressing the idea that a AC/DC brick is more efficient than a normal PSU...

Yes, an AC/DC power brick can reach >90% efficiency. It's only got one output voltage line; on laptops it's usually 14~19V. Once it goes into the laptop, it must be further converted, however, into working voltages for the CPU. memory, drives, etc. The best efficiency you'll find there in DC/DC is perhaps 90%. So by the time to correct operating voltage reaches the components, the loss is around 20% or close to it. In other words, you're looking at similar losses in the power delivery system (assuming an 80+ PSU). However, I agree that the very low power efficiency of the laptop power system is probably better than with most desktop PSUs.

I also agree with Devon's first comment -- the notebook PC is optimized to the gills to shut thing off, not provide anything unnecessary, etc.

BTW, Linus, Devon didn't just have a part in writing the iMac piece; he wrote 98% of it.

Mike, that sentence can be misinterpreted. How about "laptop power system efficiency at very low power levels is probably better..."?

yeah, ok

MikeC - Thanks for the clarification(s). All of this discussion has me thinking that the first step to lowering idle power consumption is minimizing power consumption at load, and then choosing the smallest PSU that will handle it.

Now, I realize that something like the iMac wouldn't necessarily work with an 80W PSU just because the max draw SPCR happened to record was 73W - I'm sure there are some higher peaks in power consumption (during startup?) that a watt meter wouldn't necessarily catch. But how much "headroom" do people think is needed?

Headroom --

30% is probably a good minimum number to consider. Most PSUs have a combined power rating that doesn't necessarily reflect aggregate simultaneous power delivery in real conditions. IE, they're tested into resistive dummy loads in accordance with standardized load distributions. Common sense tells me that in real PC applications, the total output power a PSU can deliver is probably somewhat less than the rated output. That in itself is reason to build in some headroom.

Then there's the question of how often you want to let any machine run at full capacity. Like how often and for how long would you take a Honda Civic up to 100mph (assuming you're on the autobahn & it can actually reach that speed) with 3 passengers in tow?

It's worth remembering that smaller power supplies are not automatically more efficient. For example, based on the SPCR article, the Phantom 500 is slightly more efficient at lower load than the 350.

I do wish there were products catering to this segment, though. I'd buy a quality, high-efficiency 150 watt fanless power supply in a second!

This is usually achievable using an AC/DC brick and a DC/DC PSU like the Pico-PSU or PW200M.

Eunos has a point. I compared efficiencies reported in SPCR's Seasonic SS-300SFD 80 Plus: Little Big PSU review and josti's The efficiency of a fanless 110w PSU kit thread, and the Seasonic actually beats the external brick kit at 40W DC output, 81.5% to 76%. Granted, that only equates to a 3.6W difference in total draw, which won't matter to most people, but every little bit counts to me.

Advantages of each solution, as I see it:

Seasonic SS-300SFD 80 Plus
- More efficient
- Less expensive

Brick & DC-DC PSU
- DC-DC PSU takes up less space in the case
- Brick (only significant heat source) is separate from the case
- Two separate components make it more flexible. For example, you could conceivably unplug the brick and connect the DC-DC PSU to a 12V battery.

Keep watching and we'll have an official NanoPSU review before long...

Do you mean the pico? Or the PW-200-M?

Sorry, yes I mean the PicoPSU

Noticed that user Hifriday reported 31W idle on the following system in How much does your system draw at idle (Poll):

ASUS Pundit P2-AE2 Barebone (Via K8M800)
Sempron 3000+ (Idle 1.1v 0.8Ghz / Load 1.2v 1.8Ghz)
1GB RAM
Samsung Notebook HDD
Slim DVD Burner
Stock case fan plus Zalman 80x15 as CPU fan

He also reported idle wattages for his Mac Minis: 21W and 25W for the G4 & Core Solo versions, respectively.

Any guesses as to whether the difference between these systems and mine is more motherboard-related or PSU-related? The Pundit has one SATA and one IDE plug, and I'm guessing the Macs have only one, whether it's SATA or IDE. All three of Hifriday's systems also have power bricks and max out at under 60W AC at full load, so they could have more efficient bricks than the one josti tested. Otherwise, I'm guessing the setups are fairly similar to mine, with onboard sound, networking, etc.

Linus regarding your CPU, noticed you are running an Athlon XP which in my experience is actually quite inefficient at idle. I believe the halt state/disconnect is not properly implemented for all socket A Athlons, if you run a software utility like S2K Disconnect, this will dramatically lower idle temps by up to 8C or more for a higher clocked chip (and assume power draw respectively). However this doesn't seem to work with all mbs and also poses stability problems even for mbs where it seems to work, probably why AMD was never able to implement it. In contrast the new A64/Semprons have significant power consumption drops at idle even without CnQ to step down voltage/clocks. I have read various articles/posts scattered through the net long ago, but sorry I don't have any links. Of course though if you're underclocking to 300 mhz and assume undervolting as well then CPU power consumption maybe already so low there is not a big difference.

In regards to the mb, I suspect that in systems like the Pundit, MacMini, and laptops, where the mb has to be designed specifically to work off the single voltage power input and have all the DC-to-DC components integrated, probably do so more efficiently than any mb designed to be powered with a regular ATX connector. The Pundi mb design is actually very clean and doesn't have a lot of components on it.

The efficiency of the DC-DC adaptors like the PicoPSU and PW200M are actually quite high. The new Pico manual (pdf) now has specific numbers listed (both rails are rated Max=6A and Peak=8A). From SPCR's power dist article, the AMD systems drew around 2-4A on these rails.

DRAW - 5V RAIL - 3.3V RAIL
1A - 86% - 85%
3A - 94% - 93%
5A - 96% - 94%
8A - 93% - 91%

For the 12V rail, I believe basically it passes through the input 12V so efficiency should be close to 100% and as we know this rail supplies the majority of the power consumed in modern systems. In my experience the PW200M is even slightly more efficient than the PicoPSU. Of course this is just the DC-DC side, the efficiency of the AC-DC brick also needs to be factored in.

When I replaced the stock AOpen XCCube 220W PSU in my Socket A system with the PicoPSU I saw a dramatic 18-20W AC power draw reduction (Stock PSU = 70-85W, PicoPSU = 52-65W) and I suspect if you are using the flex 220W that came with your case it probably is performing at similarly poor efficiency. It is possible a Seasonic PSU maybe able to match or come close to the Pico, but I have my doubts. When I finally get back my S12-430 I will try to do some comparisons, or ideally if SPCR could put up a review that would be even better.

As for Josti's Lex PW004, I do suspect it has a lower efficiency than the Pico/PW200M as it works off a higher 17.2v input meaning in addition to converting the 5v/3.3v rails it also needs to convert the 12v rail.

As for the power draws I posted for my systems, most of them are just under Prime95 and not max draw. I'll try to run RTHDRBL and CPUBURN as well as loading the HDD/DVD and post up some figures later.

Is this minimum load added to the overall power consumption or will it be replaced by the actual load once the minimum is exceeded (so for the S12 if you apply a 20W load, is the draw equal to 20+14W or just simply 20W)?

What minimum load? I specifically stated that there was no load.

I think he means the minimum power draw, ie if the Seasonic draws 14W under no load conditions, will it draw 114W when a ~100W load is applied (ie a computer) or will it simply be 100W minus efficiency losses.

Sorry my wording is not clear, the question is as Jaganath described. Actually a second question is if you apply a load <14W (say 1W) will the S12 still draw 14W or will it actually draw less?

Hifriday - To give a little more background on my system...the 37W idle figure is at 300MHz, but 1.25V is the lowest stable voltage I can get by with (I blame integrated video and maybe the PSU). I do run S2KCtl, and have seen significant drops in power consumption when clocked at my "normal' speeds (~1-1.5GHz), but it only makes a watt or two difference when clocked way down. Regardless, I can't imagine my heavily-underclocked XP needing any more than a watt or two more than your Sempron at idle. The rest has got to be motherboards and PSUs (I'm beginning to sound like a broken record here!).

Good point on the lower efficiency of josti's brick system. I also noticed that he called the brick a "230V to 17.2V converter" - maybe converting from 240V is less efficient than from 120V as well?

Question for you - does your clamp meter account for power factor? I'm wondering if those huge differences from changing the PSU on your AOpen could be due in part to active vs. passive PFC.

Devonavar - When SPCR reviews the PicoPSU, will it include efficiency testing on an AC-DC brick? I would especially like to see how 70-100W bricks compare with higher-capacity 180-220W bricks at low power draw.

ISTR that ordinary ATX power supplies run more efficiently on 230V than 120V?

FSP Green Power review

Many of SPCR's PSU reviews do mention higher efficiency working off 220v compared with 110v, but not sure if the same applies to AC/DC bricks. BTW the mains voltage where I'm at is 220v.

As for the clamp meter, it simply measures the current (amps) flowing through the wire around which it is clamped. It does not measure PF and I believe PF should not have any effect of the measured current draw (PF affects the phase of the AC line). Actually I wonder if the AC/DC brick (switching ps) has an PFC at all?

Ah, I understand what you're saying now.

I don't think you can separate the efficiency losses from the baseline power requirements because the baseline is an efficiency loss. Think of it this way: Efficiency = Output ÷ Input × 100%. With no load, output is zero, so efficiency is always 0%. But the 14W consumed with no load applied still counts as an efficiency loss. It's no different from the 14W that is lost by a power supply that is 86% efficient when it is consuming 100W from the wall. There's no such thing as "100W minus efficiency losses", plus another 14W on top of that. The 14W are part of the efficiency losses.

To answer your second question, I have never tested a power supply that drew less that what it draws with no load applied. However, it is possible that some PSU's may use a dynamic loading resistor to ensure that it starts under low load. It is plausible that a PSU with that kind of circuit might see a very slight power drop when the resistor falls out of the circuit at a higher load. Testing for this kind of behaviour would be quite difficult because it would require incrementing the load in very small increments, something that our current test setup cannot do easily.