Hacker News new | past | comments | ask | show | jobs | submit login
DIY quest for a 6 watt high-end desktop computer (tweakblogs.net)
173 points by ck2 on Sept 9, 2012 | hide | past | favorite | 31 comments



This guy should totally go work in Google's platforms power group, they would love him and give him better tools to work with.

That said, its inspiring to see folks tinkering at this level. I desoldered the floppy connector on a motherboard so it would fit into a case once and the guy who needed it was shocked that such things could be done successfully. I was pretty amused until I realized an whole bunch of people who are really bright technically have artificially limited themselves to just making changes in software. That is too bad. Old PC's are really easy to get hold of and great for practicing your rework skills, get an old Metcal iron off Ebay or somewhere, a magnifying visor, and boom go to work.

The author didn't mention a weight limit but you can passively cool the CPU with a copper heatsink on the back of that monitor, something like 9" x 21" of copper fins has a lot of surface area to work with. Might save a bit of power there too.


>The author didn't mention a weight limit but you can passively cool the CPU with a copper heatsink on the back of that monitor, something like 9" x 21" of copper fins has a lot of surface area to work with. Might save a bit of power there too.

He actually intends to do passive cooling, though I'm not certain his heatsink is going to be large enough when the CPU is running at full load for an extended period of time. You can see him discuss the casing and see the planned heatsink here: http://www.youtube.com/watch?feature=player_embedded&v=S...



Thanks for the link, it was much more detailed.


whilst it is an impressive job, the title is misleading; it's 6W idle.. 75W load.


I agree, the title is misleading. And, I see a big fan on the board.

The "low-end" Raspberry Pi, not including the monitor, draws a maximum of 3.5 Watts = 5 Volts DC x 0.7 Amps

http://www.raspberrypi.org/phpBB3/viewtopic.php?f=2&t=50...

And, RPI doesn't have a fan.


The fan looks to have been replaced in a later picture.

I've got a RPi here which is doing sterling service for several tasks, but it certainly wouldn't be satisfactory for desktop use like he appears to want - that machine is a Core i5, the two are lightyears apart! Plus the convenience factor - the RPi won't power USB devices that want more than a minimal amount of juice (an external 2.5" HDD, for example) whereas this machine could happily drive them at a temporarily increased power draw.


But it would spend a very tiny amount of time at full load?

My i3 desktop is almost always at a low of 80watts (with monitor, router, cablemodem, UPS) - full load at 150watts only happens when I fire up something with graphics (early model without GPU acceleration) and even then settles down.


Why is he not using a low-power Xeon? The 17 watt max TDP CPU is pretty amazing.

We've got a server design in house using commodity parts for a three-server cluster that's less than 60 watts per server. The whole thing (including AC/DC conversion overhead) is less than 2 amps at full draw.


These Xeons probably don't have a HD4000 graphics unit?


Ah, good point. Obviously servers don't need that. :)


> It needs to be completely silent. No fans.

You can't hear fans when they run at a very low speed. This will make your cpu cooler and use less energy


It is my experience that any mechanical component will start to make noise at some point. Better to avoid it altogether.

I don't understand why you say that using a fan would make the CPU use less energy. On the contrary, the fact that a CPU requires cooling indicates that it uses more energy than a CPU which doesn't (more energy converted to heat).


Electrical resistance is dependent on the temperature. The lower the temperature of your cpu is, the less power it uses.


this will only be true under load.

the cooling shown is not enough to manage the cpu under sustained load, and the fan is present, so either it turns on when the cpu temperature rises, or it is very infrequently under load.


I like how the battery is integrated into the computer.

I think a typical UPS converts AC to DC and then back to AC. I wonder whether this guy's design avoids those two conversion steps.


That's easy to answer in general -- if a computer runs on line power, then it has at least one AC -> DC conversion step. If it has a small, light, high-efficiency power supply that doesn't require large 50/60Hz transformers, then it converts twice:

120V AC -> 170V DC -> 20 KHz switching inverter -> 20 V DC.

So with respect to frequency it's 60 Hz -> DC -> 20 KHz -> DC. Yes, complicated, but much more efficient than the old iron-core transformer days, even though the latter involved fewer steps.

The step from 120 VAC to 170 VAC is an unavoidable increase because 120 VAC is RMS voltage, not peak voltage, and the DC conversion yields the peak value:

Peak V = RMS V * sqrt(2)

It wasn't very long ago that 170V transistors were difficult to acquire and notoriously sensitive to voltage spikes, but this problem is essentially solved.

A computer that only ever runs on batteries may have one, or no, conversions, depending on whether the battery voltage is exactly what the computer needs.

I speak as someone who has been designing electronics so long that my first designs used vacuum tubes, and whose 20 KHz inverters flew on the Space Shuttle.


Yes, he does avoid those conversions. There are two excellent posts in his blog about his custom power supply:

http://ssj3gohan.tweakblogs.net/blog/cat/1824


Great read. I wonder why the author didn't go with something like the three watt MK802 Linux mini-computer though?


I think the idea is he wants full processing power on demand while low power for the 90% of the time it's at idle.

ARM is great for idle but is still a fraction of desktop cpu computational power.

It's a couple years out of date but scroll down to the V8 and Sunspider benchmarks here:

http://vanshardware.com/2010/08/mirror-the-coming-war-arm-ve...


The new quad core ARM boxes are actually quite good. My own personal use case is different but as a Gentoo developer it's nice to see something like Firefox "only" take 3 hours to compile. Which is really good when trying to debug issues because an mx51 it takes 18-20 hours.


Incredible endeavour and i wish this catches on in a bigger way.

For now the average person can probably jump on this idea to some extent by switching to a tablet where possible. I think the ipad3 is around 4.8watts over a full charge discharge cycle with the charger unplugged after charging.


Isn't a laptop a better replacement? They are usually optimized for low energy usage, and run a full OS.

How does the energy usage of this computer compare to say, a macbook pro?


http://www.apple.com/macbook-pro/specs gives 7 hours of WiFi use and 30 days standby on a 95 Wh battery for a MacBook pro. So, it will be somewhere between 0.1-ish W and 13-ish W for idle state. I cannot find something better. For the Mac Mini, http://support.apple.com/kb/HT3468 gives 11W in idle.


FWIW I consistently get 11-13w at idle, 7 hours battery under use from my Samsung Series 7 Chronos i7. Obviously Samsung takes cues from Apple but I'm pleased at half the price.


That's for a Core i7 Retina MacBook Pro and the 13-ish W power number includes the screen. The 6W number for Fluffy2 does not.

Still an impressive hack though.


I notice his choice of motherboard conforms to the thin mini-ITX specification introduced last year by Intel.


What exactly is a voltmod?


Modifying a hardware component to feed it more or less voltage. Particularly zealous overclockers may overvolt components to allow them to run at higher clockspeeds. Particularly zealous efficiency nuts may undervolt components to trade decreased performance for decreased power consumption.


In both cases, part of the trade typically also is reliability. The farther you go outside of the spec for parts, the higher the risk that they work only sort-of. I gues that, if money wasn't a problem for this guy, he would have bought ten different CPUs and checked which one kept working at the lowest power.


Voltage modification, i.e., lowering the voltage to something. When it cannot be done in software, it can be done in hardware with wires.




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: