Fluffy2 - 5.9 Watt high-end desktop computer

Door mux op zondag 02 september 2012 23:44 - Reacties (76)
Categorie: Fluffy2, 20W all-in-one, Views: 187.762

Click here for the Dutch version of this blog

Today I am going to tell you about my newest build: Fluffy2 and how I got her as efficient as she is now. I’m only going to talk about the computer bits today, not the custom screen and other parts which will be discussed in later blogs. This will also be the post I will link to from other media (i.e. various tech forums). For those new to my blogs I would like to take the time to introduce my passion for efficiency.

I'm also on the twitters! @EfficientElec



My laboratory - no, it's not in a basement. It's on the first floor.

Mux' laboratory



What I want for my computers is low power consumption and efficiency. That last term means that I want them to achieve maximal performance while using minimal power, money and space on my desk. To accomplish this, I do not hesitate to solder the most essential parts right off the motherboard to see if that helps in it’s power consumpion. Also, to help both myself and others, I often design special electronics to make personal computers more efficient. Since I am one of the few people willing to go this far for the cause, there is not much information available about the risks and gains of these methods. With this blog, I am trying to educate people in a vast range of relevant subjects: From low power computer displays to power supply technology, an explanation of power factor correction and the worklogs of my previous low power builds: Dennis (20W Core 2 duo), the award-winning Fikki3 (8.3W Core i3) and Floppy2 (9.5W NAS), the last of which does not yet have its own blog entry. Also, I like to give my computers names so everyone knows which one I am talking about. For instance, today we are going to discuss Fluffy2, the successor of Fluffy, formerly my main computer.

http://tweakers.net/ext/f/PgUNlAe5qvdJsSIAcmEbh2Pe/full.png
A grayish 3D-render of Fluffy2

In contrast to my earlier builds, Fluffy2 is going to be an all-in-one computer: this means the computer is going to be built into the monitor. It will also be a high-end computer: The parts are among the fastest currently available, it has as much RAM as will fit in it and the monitor will be of the IPS variety. The case is a custom design with an integrated and quite massive passive cooler, and features a battery so Fluffy2 can keep working in case of a power outage or, well, if I want to take her to another room.

It's a bomb!
You might mistake this for a bomb, but it's really an 8-cell lithium ion battery that goes into Fluffy2.

The need for lower power consumption


Computers are extremely important in our modern daily life. Not just your own computer, but the computers at your office, the ones in the grocery stores, the datacenters that allow me to share this blog with you and all the computers we need to, for example, buy something online. All these computers have something in common: they are overdimensioned for their use. What I am trying to say is: None of those computers are working at full power all the time. A computer that is working at full power all the time, would be seen as slow or noisy and would be quickly replaced. The logical consequence is most actual (overdimensioned) computers are idle a lot of their time (70% idle time in datacenters and over 95% in home computers). In that idle time, the computer still uses power, but it is often impossible to turn it off to save on the electricity bill. It turns out the total power used while the computer is idle, is often more than the total power used while the computer is actually doing useful work!
Just as a quick example: right now, my computer is almost idle: My CPU-usage is less than 5%. Can I turn my computer off? Of course not, I am writing a blog! So what I want, is for the computer to use as little power as possible while it is idling. This is what I have been concentrating on for the last few years and it has yielded some nice results. Here is the idle power consumption of my computers over the past couple of years:

http://tweakers.net/ext/f/qo8QzP4krFiwotpPf4JmIUHK/full.png

How to build a low power computer


So, how does one go about making the most efficient computer in the world? The first thing you need to do is to specify exactly what you want from your computer. A lot of people would say ‘I want a fast computer that will last me 3 or 4 years’. But what is fast? And do you really mean that 3 years, or do you secretly already know you replace your PC every other year? Maybe instead you want to use it much longer, or give it to your little niece when it becomes outdated? You need to be as concrete as possible in your specification. This is a proper foundation to build on, so you won’t be tempted to change something in the foundations later on (which could collapse your whole building - to stretch this metaphor).

For instance, my own specifications for Fluffy2 were:
  • It needs to be completely silent. No fans.
  • Total power consumption including screen <20W
  • It will become an all-in-one computer
  • It needs to be altogether wireless. I want to be able to pick her up and walk away with her.
  • The monitor has to be IPS.
In a future blog, I will explain how these specifications led to the components I chose. Anyway: with your own specifications, you can check which components match your requirements. The key to your succes is: make sure your components just meet your specifications. Do not go above, do not go lower either. In my experience, at this point you will have a pretty short list of suitable components.

With this short list in hand you can make targeted low-power choices:
  • Processors: Both Intel and AMD make comparably power-efficient processors.
  • Motherboards: Intel and low-end to midrange MSI motherboards are the way to go here. Avoid Gigabyte and Asus if you're going for efficiency.
  • Memory: Do whatever you like. I've investigated this a couple of weeks ago.
  • SSDs: Avoid LSI/Sandforce and JMicron. Make sure you put as much often-requested data on your SSD so that any hard drive in your system may spin down.
  • Hard disks: Never use RAID, it has no use and has a very negative effect on your power consumption. Use as little hard drives as possible, preferably of the highest necessary capacity. One 3TB disk is more power efficient than two 1TB disks. If not more than 1TB of space is needed, go for 2.5" models.
  • GPU: If at all possible, use an integrated graphics chip, not discrete graphics cards. Most modern-day game titles, as well as basically all graphics-intensive programs can be run on IGPs nowadays at acceptable frame rates. Only use discrete graphics if they're absolutely necessary.
Advanced techniques


And now for the special Mux sauce: what do I do that makes my computers so power efficient? Apart from well thought-out specifications and of course a vested interest in low power computer components (which is something that simply requires a lot of reading and experience) I also look further than the component level in computers. I look at the components that make up PC components: the chips on the motherboard, the discrete components in a power supply. This type of information is usually not freely available, so the only way to get to know exactly how a motherboard or power supply is formed from its components is to probe and trace out everything on the boards. This results in diagrams like this showing all components on my motherboard (DQ77KB) and where those components get their power from:

Power flow diagram of DQ77KB

This is not all she wrote. Once I know how power is distributed and what components actually use power, I need to measure exactly how much power is used by what. So I need to measure the currents and voltages going through all the above. Once I know this, it is simply a matter of trying to undervolt or power down components starting from the one using the most power, all the way down to the components using the least power. In order to help me visualize which components should receive priority efficiency treatment, I make things called proportional diagrams where components that use a lot of power are really big, while power-sipping components are really small. Like this:

Proportional power consumption diagram Fluffy2 (idle)


This is super useful. It allows me to get an idea for how much leeway I still have. For instance, this diagram shows that the processor uses a bit over 1.5W of power when the computer is idle. I know for a fact that this is about as low as I can get CPU power consumption, it is a hard wall. So clearly, I won't be able to gain anything there. This puts an upper limit to the amount of power I can still save by modifying other circuits, and gives me a realistic goal to aim for. For instance, with this information in hand I can safely say that it will be extremely hard to get down to 5.0W total system power consumption.

Now, there is another interesting part to that last graphic: the brown block on top of everything called 'conversion losses'. As you may know, computers use a power supply to convert mains AC power to a voltage that is suitable for computers: 12V, 5V and 3.3V. But this is not the only 'power supply' in your computer: there are many very small power supplies in your computer that for instance convert the 12V from your power supply into about 1V for the CPU. These converters are colloquially known as VRMs or voltage regulator modules and as you may guess already: these suckers are not perfect. Every time you convert electrical power, you will lose some of that power as heat. As it turns out, a whole lot of it. Where and how these power losses occur can be visualized using a so-called Sankey diagram, like this:

Sankey-diagram Fluffy2 (idle)

In this type of diagram, the power that goes into the computer 'flows down' from above and losses incurred in power conversion are 'diverted' to the side. Only the green arrows pointing down actually eventually go into the chips in your computer to do useful stuff. If you gather up all those greem arrows, it turns out that of those 5.9 watts of mains power, only a measly 3.66W is actually used, the rest is ejected as waste heat! And what's more depressing: for computers this is actually very good. Any non-power-optimized computer, which is pretty much all other computers, use about 30W of power when idling. This is largely due to conversion losses. Only a tiny fraction actually gets used for useful activity. Even on systems using those fancy 80+ Gold or Platinum rated power supplies.

So, what is the end result?


Using the above knowledge and helpful visualizations, it was actually not that hard for me to pinpoint where I needed to modify my computer to make it as efficient as possible. It only took about a month of working in my free time to complete the task. The result?

Fluffy2's power consumption in idle and load

In this graph I recorded the power consumption before (red) and after (green) I performed my power optimizations on Fluffy2. Note - None of these optimizations impacted performance or functionality in any way!

Even though Fluffy2 was already one of the most power-efficient configurations you can buy today I still managed to chip off almost half of the idle and low-load power consumption. And the power savings are universal: even at full load, Fluffy2 used a quarter less power than before. Compared to a typical modern-day desktop which uses about 30W idle and 150W under full-load conditions, the savings are even more dramatic. I've compared my results to other people who make super-efficient computers their hobby and I think I can safely say that I have made the world's most power-efficient desktop computer by a comfortable margin!

Now, to conclude, here are a couple of nice pictures of all the modifications I've done to the motherboard. Not all of them are especially obvious, but they're nice pictures nonetheless.

http://tweakers.net/ext/f/gbbOPeGKJ8jMpTpcFkVfq8Ei/full.jpg
The computer as an 'exploded view' of its components: the motherboard, wireless networking, memory, SSD, Logitech unifying receiver and LVDS adapter board

http://tweakers.net/ext/f/5hO6OYzN5Lfz9B5RPg63po0f/full.jpg
Bottom side of the motherboard.

http://tweakers.net/ext/f/7a441mKXSamapUZfaJmEFvDh/full.jpg
CPU voltmod, because the EFI doesn't allow for voltage adjustment.

http://tweakers.net/ext/f/h3CYCBORgOQzYZT6QeEUYjc2/full.jpg
The holes for half-sized miniPCIe expansion cards are used as a way to pop a mod wire from one side of the motherboard to the other. There's also a couple of voltmods and other modwires to be seen here.

http://tweakers.net/ext/f/4PnlsOxutCmwdUJNACebURxH/full.jpg
This is where the yellow wires end up when they go through the holes.

http://tweakers.net/ext/f/bxVXPYfF5lIeI1Aetjbhlxnw/full.jpg
Because I didn't use the PCI express slot, I decided to desolder it to make accessing components on the motherboard easier.

http://tweakers.net/ext/f/jaDOr7UxAI248Xn8SBZ9P66s/full.jpg
This picture shows a lot of missing components: I desoldered a fan header, SATA ports and an LED (can you spot them?). Also, if you look very closely you can see a voltmod to the 12V regulator.

http://tweakers.net/ext/f/pQynq4MIvoLS4oHG7bkvJeDJ/full.jpg
Sometimes things do not go exactly as planned. Here I accidentally scorched a couple of components while soldering something else.

Conclusion



I hope you liked this chapter in the megaproject that is Fluffy2. But wait, this is only the beginning. This chapter is just an introduction with the fruits of my labor. In a couple of upcoming blogs I will try my best to explain in a more in-depth fashion what I did and how I did it. Also, I will publish blogs on the other aspects of Fluffy2: the UPS (battery module), true power off module and of course the case design, the last of which is already featured in a video blog. Stay tuned and if you'd like to support this project and help me do awesome power-saving work for the PC community, you can find a donation link below.

http://tweakers.net/ext/f/ARO5ytCrx0RoE07YtILpG8tg/full.png


Vorige post




https://www.paypalobjects.com/en_US/i/btn/btn_donateCC_LG.gifNo PayPal? Message me or e-mail me e-mailadres

Volgende: Fluffy2 - 5.9W high-end desktop computer 09-'12 Fluffy2 - 5.9W high-end desktop computer
Volgende: Mux' vragenuurtje 17/8/2012: hoe zuinig is geheugen? 08-'12 Mux' vragenuurtje 17/8/2012: hoe zuinig is geheugen?

Reacties


Door Blog reader, zondag 09 september 2012 18:21

1) It would be nice if you had a parts list, e.g. which processor model, SSD model, motherboard model numbers.

2) Could you take some pictures of the finished product?

Door Tweakers user mux, zondag 09 september 2012 18:54

Sorry, the project isn't finished yet. This is one of the biggest parts in the project, but e.g. the casing isn't done yet. I'll try to get a link on hackaday once the whole thing is finished and characterized. You'll get lots of pictures of the end result then.

As for the parts:
Intel Core i5-3570K @ stock speed
Intel DQ77KB motherboard
2x8GB Crucial DDR3-1333 (also stock speed)
Intel Ultimate-N Wifi-link 5300
MyDigitalSSD 64GB
Logitech MK701 wireless keyboard set
Custom LVDS/LED power supply board
Custom UPS-board and battery
Custom casing
Custom cooling
Panel from LG IPS231P

Door Fox8, zondag 09 september 2012 19:15

Components (as far as I can get it from the pictures and text):

- Motherboard Intel DQ77KB
- CPU Intel Core i5 3750K
- DDR3 SO-DIMM Corsair 2x8gb (unknown frequency and model, probably Value 1600Mhz)
- Wireless mini-PCIe Intel 4965AGN
- SSD 64gb mini PCIe (unknown manufacturer)

Door Fox8, zondag 09 september 2012 19:17

Oh well for some reason my comment appeared too late :P delete it if you want.

Great project mux you just inspired my next build!

Door Tweakers user mux, zondag 09 september 2012 19:23

I only delete spam and inflammatory stuff :D thanks for the effort anyway

Door Tweakers user mux, zondag 09 september 2012 19:26

By the way, about the wireless card: the 4965AGN wireless card was a temporary way to get MIMO capable wifi up and running on the computer while I was waiting on the (proper, half-size) Ultimate-N 5300 card to arrive. That's why in the photos you can see a 4965AGN but in the list above I speak of an Ultimate-N 5300

[Reactie gewijzigd op zondag 09 september 2012 19:26]


Door Pepijn de Vos, zondag 09 september 2012 20:38

Gaaf man, ik zou best nog wel meer willen weten over waar al die draadjes, of juist het gebrek aan draadjes voor zijn.

Hoe is het verbruik in vergelijking met gelijkwaardige laptops? Ik zou toch denken dat fabrikanten er daar alles aan doem om zo zuinig mogelijk te zijn.

Ik zit zelf te denken of het mogelijk is om een e-ink als secundair scherm te hebben. Het blijkt dat je scherm best een groot deel van je energie verbruik is.

Lijkt er alleen om dat je die in grote getallen moet bestellen bij een fabrikant. IdeeŽn?

Door Tweakers user mux, zondag 09 september 2012 21:49

In het volgende blog ga ik, zodra ik daaraan toe kom, uitleggen waar alles precies voor dient en hoe mijn testmethodologie was.

Het verbruik is ongeveer gelijk ŗ ietsjes lager dan alle laptops. Het dichtste bij komen ultrabooks momenteel, die ongeveer 5W DC (~0.5W minder dan Fluffy2) verbruiken in idle, maar daartegenover staat dat er ongeveer 4x zo trage processoren in zitten. Ik denk dat ik met een Celeron G530 ook wel op zo'n verbruik moet zitten. Dat is overigens excl. scherm.

E-ink is op de desktop IMO niet echt een geweldig alternatief. Ja, het is het summum qua verbruik, maar je levert er een hoop functionaliteit voor in: je kunt niet normaal video kijken, kleuren zijn erg slecht (als je al een kleurenscherm kiest) en het ding is alsnog heel duur voor het oppervlak dat je krijgt. Er bestaat bovendien nog niet zoiets als een 20" e-ink scherm, de grootste OEM-panelen zijn vziw 12".

Dan liever 3-4W in een TN-paneel steken, eerlijk gezegd.

Door Criacao, zondag 09 september 2012 22:11

Fantastic project this Fluffy2! I look forward to the next updates!

Door Nico, maandag 10 september 2012 05:11

Hi.
How would your work compare to the average laptop (let's say Macbook Pro, around 90Whr battery for around 10hrs idling, screen on) ?

Nico

Door Tweakers user mux, maandag 10 september 2012 09:27

It's about the same. Ultrabooks have a slightly lower power consumption, but also less powerful components.

Door sunna, maandag 10 september 2012 09:47

Nico schreef op maandag 10 september 2012 @ 05:11:
How would your work compare to the average laptop (let's say Macbook Pro, around 90Whr battery for around 10hrs idling, screen on) ?
Well, the CPU in fluffy2 seems to be about 50% faster than the one in a MBP, for a start. Also, apple.com states 7hrs idling for the MBP, 90Whr/7hrs are about 13W idling.

Door T.Terlemez, maandag 10 september 2012 10:26

Very nice. Thanks for sharing.

Door Pepijn de Vos, maandag 10 september 2012 12:20

mux schreef op zondag 09 september 2012 @ 21:49:E-ink is op de desktop IMO niet echt een geweldig alternatief. Ja, het is het summum qua verbruik, maar je levert er een hoop functionaliteit voor in: je kunt niet normaal video kijken, kleuren zijn erg slecht (als je al een kleurenscherm kiest) en het ding is alsnog heel duur voor het oppervlak dat je krijgt. Er bestaat bovendien nog niet zoiets als een 20" e-ink scherm, de grootste OEM-panelen zijn vziw 12".

Dan liever 3-4W in een TN-paneel steken, eerlijk gezegd.
Als primair scherm lijk het me ook niet ideaal, maar het lijkt me heerlijk om met een tiling window manager, VIM en een web browser te lezen en schrijven. Zegmaar een betere Kindle aan je desktop/laptop. Ondertussen kan je primaire scherm de helft van de tijd uit.

Jammer alleen dat je nit zomaar zo'n e-ink scherm kan kopen, alleen dure dev kits.

Door Jurriaan de Vos, maandag 10 september 2012 12:52

Wow, super chill!
Ik ben ook wel redelijk into hardware, maar zů ver ben ik nog nooit gegaan. :-)

Toen ik een oud-lerares(duurzaamheids-fan) vertelde over dit artikel, wees zij me op een jaarlijkse wedstrijd van de ASN bank voor duurzame ideeŽn en producten, waar een aantal prijzen worden vergeven om startende ideeŽn een financieel opstapje te geven: http://www.njr.nl/Nieuws/Nieuws-NJR/ASN-Bank-organiseert-duurzame-wedstrijd.html
Mocht je dus ambities hebben om het meer te maken dan een hobby project en het op de markt te zetten, dan is het misschien een optie om een mobo te ontwerpen, met de goede embedded chips, voltage controllers, en een custom, al dan niet build-in PicoPSU(Dennis).

Een beetje zoals de Cubieboard en de RaspberryPi, maar dan wat higher-end.

Nu ik er aan denk, Intel heeft geŽxperimenteerd met servers koelen met minerale olie, en heeft zo de energie die nodig is voor het koelen van 50% teruggebracht naar 2-3%: http://www.tomshardware.com/news/Server-Cooling-Hardware-mineral-oil,17348.html#xtor=RSS-999

Mocht je het leuk vinden om te sparren over vermarkten, koeling, whatsoever, mail gerust, of laat een comment op mijn website achter of iets dergelijks. Ik draag dit soort projecten "een warm hart toe" zoals dat heel lief heet. :-)

Jurriaan

Door Tweakers user mux, maandag 10 september 2012 13:01

Oh wow, dat is een leuke prijs om aan mee te doen. Eerder heb ik al eens meegedaan en een prijs gewonnen met de Universiteitsfonds Delft-Cofely Energy Efficiency-prijs (linkje) en dat heeft me flink verder geholpen, dus ik zal zeker een gooi doen naar die ASN bank-prijs. Ik ben zo te zien nog nŤt op tijd...

Ik ben ook altijd op zoek naar mensen die met me willen nadenken over het verder verspreiden van... tja... de dingen die ik doe. Ik zal straks eens een mailtje sturen.

Door Davos, maandag 10 september 2012 16:37

I like this very much. Modding for too long has been about 1000W power supplies and over clocking. Efficiency is a much smarter direction. I hope this sort of hardware hacking becomes really popular. Then instead of making overclock-ready motherboards for enthusiasts, we will see power saving options.
To be fair, the industry has been doing this already. Technologies like speed step are one example.

Door Tweakers user mux, maandag 10 september 2012 16:42

Definitely, since Pentium M processors have been steadily improving on the idle power front to the point where nowadays they use almost 2 orders of magnitude less power then at the end of the Pentium 4-era. I couldn't make a 5.9W computer with that generation of hardware. But still, it's pretty cool to realize that on top of all that power efficiency that chip manufacturers already realized, there's another 50%-or-so to be optimized away by improving power delivery.

Door thx1200, maandag 10 september 2012 17:16

Amazing what you have done, but I couldn't let this comment pass.

"Never use RAID, it has no use and has a very negative effect on your power consumption"

I hope what you meant was "Never use RAID-0." Because with SSD, it's usefulness is indeed questionable, but other RAID levels are extremely useful if you want to provide fault tolerance, even if it is less power efficient.

Door Tweakers user mux, maandag 10 september 2012 17:20

Not necessarily. RAID is pretty damn inefficient when it comes to fault tolerance, because at the same time that you're adding fault tolerance (which, don't forget, overlaps in functionality with already built-in fault tolerance measures like integrity checks and protocol-level error handling) you're also increasing the likelihood of failure by adding complexity to the system (in the form of an extra disk, the necessary software and hardware routines that enable RAID, etc.). RAID-1 mathematically is even less reliable than a single disk, although in practice it's a little bit better.

The only reason to run RAID ever is to decrease your downtime, because even though you have a lot more failures, a single failure is not a complete system failure. It is not a way to improve data integrity though.

[Reactie gewijzigd op maandag 10 september 2012 17:40]


Door xanonx, dinsdag 11 september 2012 14:53

"It is not a way to improve data integrity though"

I do believe that is the very definition of RAID: "resiliency, performance, and capacity"
Above all, what you're doing with RAID is a form of hardware assurance; if a disk dies, the information is still readily available. Granted electro-mechanical failures may not be a problem for SSD's, who knows what sorts of Alpha particles and crazy ex-lovers your harddrive might encounter.

As for RAID 0'ing SSD's, there's definitly a noticable diference in performance for anything involving HD video. Some sauce: http://www.xbitlabs.com/a...n-hyperx-ssd-raid0_6.html. And what does a pleeb really need a high-end machine for other than HD video? While I spend hours emulating FPGA's in my basement, very few of my friends can say the same.

Door petersys, woensdag 12 september 2012 01:59

Hi,

nice to find and read your blog (partly via Google Translator only).
Could you be a bit more specific about the mods.
I would do some slightly modifcation too. I.e remove unwanted LM317 (for TPM) or similar hacks like rerouting powerlines as you did using the fat yellow wires.

Would be a great pleasure to see a guide here.

Thanks!

Door Tweakers user mux, woensdag 12 september 2012 08:14

There will be a blog post about this soon, be sure to RSS or bookmark me so you don't miss it.

Door Matthew, woensdag 12 september 2012 16:15

You, sir, are an absolute genius and are able to do things that I wish I could even begin to contemplate. It is a great shame that so much of modern day life is not designed with such efficiency in mind.

It reminds me of the rather excellent podcast episode of 99% Invisible about how ubiquitous and cheap energy has meant that modern objects are poorly designed in respect of their efficiency.

http://99percentinvisible...undesigned-but-still-evil

Door Lorride, woensdag 12 september 2012 19:16

Cool project you have going here! You do touch upon something I have been angry over lately when building my own new and power efficient work station computer. Especially during idle, where in my tiny PC the terrible PSU is wasting away soo much power. Which is a big problem all over the board, now that Mini-itx is getting more foot hold and the old "Shuttle Cube" or "Aopen Cube" is getting a bit more competition we still see so many severely overkill PSU's and bad characteristics at low power.

Tips if you wanna design some small very power efficient circuits for specific voltage needs and or replacement of shitty LM3x7, 78xx or 79xx:
LM2675 <- Expensive, but simple and very efficient. Simple to use, good for rapid development.

AS1341<- Cheap, but requires a bit of components, hard to solder given its package, highly customizable.

MC33063ADG <- Cheap, not too many components needed and highly customizable.

I have used 1. and the 3. one in my designs, but have no experience with the 2. I highly recommend looking them up ;)

Door Tweakers user mux, woensdag 12 september 2012 22:52

Thanks for the heads up, but really... there are sooooooooo many chips available.

I'm actually a professional electronic design engineer who specializes in power electronics: all day, every day I do nothing but make dc/dc converters. I know my stuff! I wouldn't say I know everything, but I certainly know a lot more than your suggestions. Not trying to be condescending by the way, I appreciate the thought.

I'm actually in the process of trying to find funding to design my own PSU from scratch, with efficiency in mind. It's called MADPSU and by the end of the year I'm sure you will hear a lot more about it!

Door Michael Adams, donderdag 13 september 2012 02:41

Great article: found it via the grano.la power management folks. I would love to see what you could do with an AMD A6 or A8 chip: I've built some work machines that idle around 40W (if I'm recalling correctly) on those; the E350 boards are low-mid range, and idle as low as 17W stock.

Door Tweakers user mux, donderdag 13 september 2012 09:05

I've seen reports out and about of A8-powered machines that get very low idle power consumption as well (as in: near 10W). Same goes for Llano- and Brazos-powered stuff. Problem is: I went for efficiency here, i.e. high performance at low power. AMD gets the low power part, but their performance/watt is about twice to three times as bad as Intel's. That's just not acceptable for a machine that is going to be regularly used for video rendering/misc. CPU intensive tasks.

Door Lorride, donderdag 13 september 2012 10:09

mux schreef op woensdag 12 september 2012 @ 22:52:
Thanks for the heads up, but really... there are sooooooooo many chips available.
snip...
Hey, not at all =)

It's not always easy to spot peoples specialization within professions through a bit of a blog, even though the tools(oscilloscopes for example) highly suggested someone with a bit more experience than the everyday PSU tinkerer. ^^

MADPSU, great, I am absolutely gonna keep that in mind! ;)

Door Tweakers user mux, donderdag 13 september 2012 10:15

About MADPSU, that italic text at the top of the page: that's about MADPSU. I'm running for a §10,000 prize which will fund the further development, testing and all the necessary steps to sell a commercial grade MADPSU. In the upcoming months (starting Sep. 18th, ending Nov. 8th) I need as many people as possible to vote on my project so I can win that prize. I've got some stiff competition so I can use all the votes I can get. If you're interested I'd love also for people outside the netherlands to vote for me.

In return the entire project will be open source, so everybody gets CE/FCC/UL approved schematics, layouts and a huge mess of blogs explaining it in minute detail.

Door Lorride, donderdag 13 september 2012 10:41

You sure have a lot of competition, and a lot of votes already ^^

However I think this is truly a part of desktop computing that really needs a kick to improve, so I will give you a vote on the 18.

Also since you mention it to be open-source with documentation, then I may be able to use it in some products that I am working on.

Door Tweakers user mux, donderdag 13 september 2012 10:43

Sure, although I am debating putting in a beer-ware license (i.e. if you're going to sell your stuff commercially, drop in once in a while and buy me a beerlemonade (i don't like to drink)).

Door Lorride, donderdag 13 september 2012 11:06

Depending on how it turns out, I would not mind some kind of royalty per unit, but if you insist on lemonade that can be arranged as well ;)

Door WindWalker, vrijdag 14 september 2012 01:38

Hello,

What is the brand/model of your wattmeter/power metter/energy meter?


Thank you

Door Tweakers user mux, vrijdag 14 september 2012 08:24

I use a Voltcraft Energy Logger 3500 and an Elcontrol Microvip mk1.1 industrial meter with x10 current clamp.

On the dc side: Fluke 111, 183 and 185

[Reactie gewijzigd op vrijdag 14 september 2012 08:25]


Door kaskode, dinsdag 18 september 2012 12:06

mux,

great to see your update, and I'll be curious about what mods you performed. Couple of questions / comments here:

1. Your unmodded power of 11.6 W w/o video output and 16.9 W with seems a tad high, considering you used a board with a mobile chipset and on-board power conversion. There are a couple of reports out there of folks who achieved a similar or slightly lower draw with regular boards and a pico psu, both with series 6 and 7 chipsets.

Then there is the guy over at hardforum.com who claims to have achieved 11 W with video output active and GLAN connected, and down to 8.8 W w/o video and fans but GLAN still connected. He used a DH77DF, an i5-3470S, MW power brick and pico 90W XLP psu. We don't know how accurate his power meter (GT-pm-04) is, though.

Then you wrote at hardwareluxx that you have an G620 system that consumes 9.4 W with screen on. Can you give more details about the configurations and mods applied if any?

2. You claim repeatedly that a Celeron or Pentium Sandy Bridge CPU will give you lower idle power. From the reviews I have seen where folks tested different CPUs in the same setup, the higher powered i3/i5 seemed to have the same or even lower idle power than the cheopo or T version CPUs.

3. As for power meters, there was a test in c't magazine and another in Elektor about two years ago (maybe I can undig them). Apparently, of the readily availble meters only the ELV series and a cheap one that can be bought at Reichelt for 10 § are any good at low powers. The Voltcraft all had issues at low power.

4. LG has just released the 4 series of displays with AH-IPS panels (I have no idea why they would release a 4 series after the 5). The IPS224V is rated at 25 W as opposed to 31 W for the IPS225V or 35 W for the IPS231P. I have a 224 on the way and will report on the measurements soon.

5. I would love to hear more about the LVDS adapter you have designed. From what I know from replacing laptop screens, the LVDS connector to the panel is standardized (at least for anything built after 2007), but the backlighting is not.

Best

Eric

Door kaskode, dinsdag 18 september 2012 12:29

one more question, where is the link where we can vote for your mad-psu project?

Door Tweakers user mux, dinsdag 18 september 2012 12:36

Hi kaskode :)
kaskode schreef op dinsdag 18 september 2012 @ 12:06:
mux,

great to see your update, and I'll be curious about what mods you performed. Couple of questions / comments here:

1. Your unmodded power of 11.6 W w/o video output and 16.9 W with seems a tad high, considering you used a board with a mobile chipset and on-board power conversion.
It's not a mobile chipset, it's desktop. And I am using a bit more power initially because of the wifi card, wireless desktop (and need for the USB controller to be on at all times), less-than-ideal SSD and high-end CPU. The power consumption without all these things should be just over 10W (I've seen reports of other people with a DQ77KB).
Then you wrote at hardwareluxx that you have an G620 system that consumes 9.4 W with screen on. Can you give more details about the configurations and mods applied if any?
That system is called Floppy2 (my NAS), the 9.4W is with screen OFF, but including the hard drive. It also has a built-in battery. I'll write or tape a blog about that at some point in the future
2. You claim repeatedly that a Celeron or Pentium Sandy Bridge CPU will give you lower idle power. From the reviews I have seen where folks tested different CPUs in the same setup, the higher powered i3/i5 seemed to have the same or even lower idle power than the cheopo or T version CPUs.
I've tested it with a G530, it's about 1W less. Most people don't measure correctly. I've seen people forgetting to turn on Intel C6 on multiple occasions.
3. As for power meters, there was a test in c't magazine and another in Elektor about two years ago (maybe I can undig them). Apparently, of the readily availble meters only the ELV series and a cheap one that can be bought at Reichelt for 10 § are any good at low powers. The Voltcraft all had issues at low power.
This is a very sticky matter, because those magazines are basically wrong. They just measured (or even copied other people's measurements) the absolute difference in power measurements from those devices and a reference industrial meter and built their conclusions from there. That's wrong. The defining feature of a good energy meter is that it can do hardware integration and has true rms capability with a reasonable measurement bandwidth. Nothing else matters. Any linear or offset error can be calibrated out.

The voltcraft energy logger (just the one, not any of their other series) has an Arch Meter chip with a real DSP for integration and ~10kHz bandwidth, which is about as high as you need to go to get good measurements. Like the Kill-A-Watt it has quite a significant offset (and quite some variation between models) which likely points to their use of copper strip measurement resistors instead of proper low-tempco ones. But aside from that they're perfectly fine meters. In the Netherlands it's one of the very few true rms meters you can buy.

Besides the Voltcraft I also use an industrial quality Elcontrol Microvip mk1.1 with updated firmware and x10 current probe, which has a guaranteed max 0.2W error over temperature.
4. LG has just released the 4 series of displays with AH-IPS panels (I have no idea why they would release a 4 series after the 5). The IPS224V is rated at 25 W as opposed to 31 W for the IPS225V or 35 W for the IPS231P. I have a 224 on the way and will report on the measurements soon.
the IPS231P - and for that matter most LG IPS screens - aren't actually that good when it comes to power efficiency. But they have such a good price point that I took one of them. I'm sure they'll be able to shave off more than 10W in the future, even at 25W they're only middle-of-the-road IPS LED backlit screens.
5. I would love to hear more about the LVDS adapter you have designed. From what I know from replacing laptop screens, the LVDS connector to the panel is standardized (at least for anything built after 2007), but the backlighting is not.
I'm still hesitant to start experimenting with it... I've only got one shot at it. Give me luck!

Door Tweakers user mux, dinsdag 18 september 2012 12:37

kaskode schreef op dinsdag 18 september 2012 @ 12:29:
one more question, where is the link where we can vote for your mad-psu project?
Unfortunately voting is only open for dutch IP addresses. If you have any dutch friends, let them visit:

http://voordewereldvanmor...ciente_computeronderdelen

Door kaskode, dinsdag 18 september 2012 13:43

Oh, that's why I didn't even see the link when you mentioned it previously :(

Regarding the guy with the DH77DF and GT-PM-04 power meter, I found two search hits on this power meter. One is a Swiss review that contains few detailed data:
http://www.topten.ch/uplo...messgeraete-SAFE-2011.pdf - but the meter does not appear to be too bad.

The other is an old thread on Meisterkuehler. Apparently, one guy measured it against the well-respected ELV 600 and got higher readings on the GT-PM-04, the other one got roughly matching readings.

I agree that when you can do your own calibration than bandwidth is all that counts.

However, when we want to assess other peoples measurents, we also need to know what kind of systematic and random errors to expect for a certain model.


Curious thing about the i5 vs celeron. Maybe the additional cores still have some residual power draw in idle. On the other hand, the guy with the DH77DF used a very similar CPU to yours.

Well, I got myself a used dq77kb and i3-2100 on ebay, so I should be able to report my numbers shortly.

Door Tweakers user mux, dinsdag 18 september 2012 13:54

The biggest difference between CPUs nowadays is simply die size. There is still about 0.5-1W of leakage in mostly the cache. So if you take G530 (which is a cut-down 2-core sandy bridge part, about 100mm2) or an i3-2100 (same die size) you shouldn't see much difference. But comparing an ivy bridge GT2 quadcore and a dualcore sandy bridge (about 2x die size difference) you should see the idle power consumption change.

That being said, I got just as much power savings switching to Windows 8 instead of Windows 7 as I got from switching processors. So there's still a lot to be won on the software side.

Door kaskode, dinsdag 18 september 2012 14:34

I agree with the leakage part.

Now the windows part is simply amazing. All reports I have seen seem to agree that Win8 offers no power or performance advantages, but as you suggest, maybe they do not measure precisely enough. Still, it is amazing that a less than mature Windows that even relies on lots of Win7 drivers should already be better.

Now all we need to find is a low power HD+ or better touch screen...

Door kaskode, dinsdag 18 september 2012 14:59

Just out of curiosity, and maybe, since you have looked more into VRM configurations of Intel CPUs, you are in a position to answer it:

Wouldn't the easiest way to beat the leakage issue be not to power unused sections? In the case of the Pentium or Celeron, this would mean not putting the bond wire in place that powers parts of the cache or even the second core in the case of single core Celerrons. For the quad cores, this would mean using a separate power supply pin for one core that needs to stay semi-awake vs. the others.

Door Tweakers user mux, dinsdag 18 september 2012 16:19

kaskode schreef op dinsdag 18 september 2012 @ 14:34:
Now the windows part is simply amazing. All reports I have seen seem to agree that Win8 offers no power or performance advantages, but as you suggest, maybe they do not measure precisely enough. Still, it is amazing that a less than mature Windows that even relies on lots of Win7 drivers should already be better.
Wait, what? Where did you read that, because there is a pretty dramatic difference between Windows 7 and 8. For one, what seems to be the major improvement is that the Windows 8 OS tick rate is much lower, around 20-30Hz (down from 60ish). On RT devices it's said to be adaptive down to 1Hz. Secondly, in idle Windows 8 does task deferring so that tasks become concentrated around a critical interrupt or OS tick, which leaves the IA cores in ACPI C3 for much longer than in Windows 7.
kaskode schreef op dinsdag 18 september 2012 @ 14:59:
Just out of curiosity, and maybe, since you have looked more into VRM configurations of Intel CPUs, you are in a position to answer it:

Wouldn't the easiest way to beat the leakage issue be not to power unused sections? In the case of the Pentium or Celeron, this would mean not putting the bond wire in place that powers parts of the cache or even the second core in the case of single core Celerrons. For the quad cores, this would mean using a separate power supply pin for one core that needs to stay semi-awake vs. the others.
They do that, it's called power gating. The problem is that you can't powergate SRAM, and SRAM has the biggest area (although leakage per square mm is much lower than, say, execution units). The only point at which the big caches are completely flushed and powered down is deep sleep which is a state you don't necessarily want to be in in S0 (because it takes a processor 20+ ms to wake up from that state).

Door kaskode, dinsdag 18 september 2012 16:48

This is where:
http://www.tomshardware.d...,testberichte-241095.html

What you write seems intriguing. It's just that the new GUI seems to be a PITA if you don't have a touchscreen, from what I've read.

Does it really hurt for the CPU to be in S0 if it takes only 20+ ms to wake up? This is much shorter than the time it takes the HDD to spin back up and the screen to produce a picture. I suspect this is not used because Windows takes much longer (more like a couple of seconds) to wake up from this state.

Door kaskode, dinsdag 18 september 2012 16:56

actually, just reread the article, and they don't even talk about power consumption - so I guess I must have picked that part up elsewhere

Door Tweakers user mux, dinsdag 18 september 2012 16:59

kaskode schreef op dinsdag 18 september 2012 @ 16:48:
This is where:
http://www.tomshardware.d...,testberichte-241095.html

What you write seems intriguing. It's just that the new GUI seems to be a PITA if you don't have a touchscreen, from what I've read.
First thing I did was remove hot corners, install classic shell and and direct-to-desktop and you won't be bugged by the horrible UI ever again. It seems like classic shell is well-behaved and doesn't induce any extra power consumption, the remove hot corners thing is just a couple of registry entries, so it's safe to use.
Does it really hurt for the CPU to be in S0 if it takes only 20+ ms to wake up?
The problem is that during those 20ms it uses power but doesn't actually do anything useful. If your OS doesn't properly manage state transitions the CPU might be constantly bouncing between C-states. The only way to save energy with those really deep sleeping states is to have them last as long as possible.
This is much shorter than the time it takes the HDD to spin back up and the screen to produce a picture. I suspect this is not used because Windows takes much longer (more like a couple of seconds) to wake up from this state.
S0 means that for the OS and user the computer still seems 'on'. It is not like standby or something. So during S0 the screen is probably already on and the hard drive is probably operational.

Door Tweakers user mux, dinsdag 18 september 2012 17:00

kaskode schreef op dinsdag 18 september 2012 @ 16:56:
actually, just reread the article, and they don't even talk about power consumption - so I guess I must have picked that part up elsewhere
Oh sorry, I didn't actually react on that. Nope, there's no mention of power consumption, only performance. I can't actually say anything about that.

Door Lorride, woensdag 19 september 2012 11:59

Unfortunately voting is only open for dutch IP addresses. If you have any dutch friends, let them visit:

http://voordewereldvanmor...ciente_computeronderdelen
Are you entirely sure? I was able to vote, saw the vote number increase and when I tried voting again it said that I already had voted. =)
I am located in Norway, IP shows it as well.

Door Tweakers user mux, woensdag 19 september 2012 12:05

Interesting... The rules say it's just for dutch IP addresses.

Door Oliver, woensdag 19 september 2012 16:27

Daaarrnn 5,9W....this is just unbelievable :o Goog job!! _/-\o_

BTW: What was the original idle power consumption without these soldering hacks?

Door Tweakers user mux, woensdag 19 september 2012 16:32

You can find that under 'So, what is the end result?'

Door www.requisiteclothing.com, maandag 08 oktober 2012 13:38

We’re a group of volunteers and starting a new scheme in our community. Your website offered us with useful information to work on. You have done an impressive process and our whole group can be grateful to you.

Door Gabe, zondag 21 oktober 2012 07:12

I'm curious what idle power gains would be had going to Haswell:
http://images.anandtech.c...ell/Architecture/idle.jpg
(Yes, unscientific graph under dubious test conditions, but it's all Intel released). Looks like Haswell will be bringing incredible efficiency improvements across the board, too.

In the meanwhile, I'm curious why you didn't start off with a strong mobile board/CPU like some other commenters wondered. I'm sure a number of folks would argue that the i7-3667U is sufficiently "high end" -- yet takes 1/5 the power of the 3570K (and delivers somewhere around the order of 2/3+ the performance for less heavily threaded applications). If that's still too underperforming to quality for "high end", then for sure the i7-3840QM would handily outperform the i5-3570K at just over half the power cost.

In any event, I think there is some real efficiency to be gained in the coming year are in the Haswell U sector with "full performance" under 20W max TDP and enhanced S0ix for idle, along with PSR for the panel. Coupled with Intel's ramping up its pressure on OEM component producers and motherboard power-gating crackdowns, we can expect to squeeze even more efficiency out soon enough...

Door Tweakers user mux, vrijdag 26 oktober 2012 10:46

For starters, yes, Haswell is going to mean big business for mobile devices. Desktops... eeh, not so much. We've already gotten to sub-2W idle figures for desktop processors (Fluffy2 is the living, breathing, but actually not living nor breathing because it's a computer, example). Nobody is going to care about idle power gains anymore here. Desktop haswell is going to be higher TDP, higher performance ivy bridge. I'd highly recommend listening to the Anandtech podcast and their opinions on the platform, very informative.

But now the meat of your comment: using mobile parts. If you really think about it, why would you? They're exactly the same dies as desktop parts, clocked lower, slightly more aggressive power settings. Different footprint. But they are basically the same. At much higher production and retail cost. Getting an i7-3667U means shelling out almost 400 bucks on the processor alone. And you can get a 2.5x higher performing i5-3570K for $230. So why, ever, buy a MoDT system?

Because it's the same chip, the same power saving techniques, the same idle power use. Using a desktop chip simply doesn't mean it *has* to use more power. We let it use more power because we can, not because we have to. So if I would want to have an i7-3667U.... I would rather buy an i5-3570K, underclock it and disable two cores. As stupid as it sounds, it's cheaper and frankly better because desktop systems are much more flexible.

And also, don't forget that I'm not just building power efficient systems at any cost. I am building power efficient systems that have ridiculously low lifetime cost. Fluffy2 costs, in hardware, materials and manufacturing, just shy of 700 euros or $900. That's with an IPS screen, built-in battery and UPS, highend processor, tons of RAM, quick SSD, the whole shabang. Try buying a laptop with similar power efficiency *and* performance for that money, they don't exist. Or any other computer.

Door finn777, zondag 04 november 2012 12:49

Good afternoon. And where you acquired adapter LVDS 40 pin LVDS 30 pin or its number. Thanks.

Door Tweakers user mux, zondag 04 november 2012 13:17

I designed it myself, when I am happy with the design (it isn't fully tested yet) I will put it up for sale in my webshop, which also isn't quite there yet :P

If you really want one you can contact me on e.nijssen@gmail.com and you can buy one directly from me

Door Valentin, woensdag 07 november 2012 07:41

Mux,
Thanks million for your website, it's tryly amazing what you are achieving there. I have in mind to build an ultra efficient low power pc, but never thought that we could even go lower than what stock power consumption would give us.
Hope you'll write up a detail guide in the future if we want to build something similar.
Luckily, I got some dutch family :)

Door john3voltas, zaterdag 10 november 2012 23:21

@mux
Thanks a lot for sharing your findings with us. Here at home we got very excited with what you wrote in this Fluffy2 blog entry.
I have always dreamed of assembling an HTPC that can either run XBMC and run some great Linux games (FlightGear, TORCS or Speed-Dreams, etc) or even reboot into Win7.
I just never tried it because of the electricity bill. You know, we're Portuguese...
This HTPC wouldn't be always ON. I have a small ARM plug computer (Seagate Dockstar) that consumes around 8w that does a bit of home server (file server, print server, etc).
And it can also be used as a PVR/DVR backend. I could use it to store all my files and use the HTPC to read the files.
I've never been good with a soldering iron, which means that I would never mod my PC motherboard as you did. But Fluffy2's hardware used 17w idle with screen ON which are some very respectable figures.
Which leads me to some questions:
- would you think that Fluffy2's hardware is capable of running some games like FlightGear?
- the non-modded version of Fluffy2 got 17w in idle and 100w in full load. I think watching an HD movie (1080p) wouldn't take Fluffy2 to full load, right? How much would non-modded Fluffy2 consume while watching an 1080p or while playing a game?
- would you help me choose the best and parts for my HTPC? Best as in low power consumption and cheap? I'm talking about mainboard (needs gigabit ethernet, HDMI and optical sound connector for connecting to a home theater sound system that I have), i3 or i5 cpu with cooler (a quiet one), memory modules 2 or 4GB. OS would then be booted off of a big CF card or SDD. No need for bluetooth or WIFI.
Thanks a bunch.
Cheers

Door john3voltas, zaterdag 10 november 2012 23:25

Oh! I almost forgot!
I also would need advise on a power supply. No need for UPS ;).
And if you know any HTPC computer case that you can recommend... ;)
Cheers

Door Tweakers user mux, zondag 11 november 2012 01:01

Send me an email: e (dot) nijssen (at) gmail (dot) com

Door FNW, dinsdag 11 december 2012 09:49

Hi Mux,
just one word: AWESOME !
I just found your amazing article while searching for an ultra-efficient future replacement for my current "server" machine. It's actually a ULV laptop I try to get completely powered by solar energy (which I have to harvest from inside my living room, unfortunately - the problem of living in an apartment).
As I am eagerly willing to rebuild your hardware mods myself (soldering experience existent), it would be great if you could provide some more details about the exact points of modification. If you used any additional parts (e.g. new voltage regulators - hard to tell from the photos), I would be glad to have them listed, too.

Thanks A LOT for starting this amazing project !

FNW

Door -, vrijdag 28 december 2012 03:21

The i5-3570K is not shown as supported on the DQ77KB support list.....

Did it boot?

Door Tweakers user mux, vrijdag 28 december 2012 09:33

It only boots with the previous BIOS version.

Door Naam, vrijdag 28 december 2012 10:08

Well that sucks as i wanted to use a i5 3750T (45w TDP) and thats not supported either.

Any idea if the older bios supported it?

Door Tweakers user mux, vrijdag 28 december 2012 10:18

Probably, yes. The earliest BIOS did not have microcodes for any of the i5 35xxs (or i7s) so it will accept those processors just fine.

Door -, vrijdag 28 december 2012 10:30

Thankyou. :)

Wonder why they removed support for so many processors.

It now supports less than a cheap H61 board.

Door Roy Batty, vrijdag 04 januari 2013 14:28

Dear Mux,

First of all, Stellar work.
If only all motherboard manufacturers took a leaf out of your book.
Secondly, what you have done is so AWESOME is that it is off the freaking scale.
[Awesomeness factor unable to calculate as the awesomeness meter may be broken or the sheer awesomeness has overloaded the meter.]

I have only one question to ask and that is regarding the GPU.
Do you plan on releasing any plans or schematics for a similar system but with an i7 processor and room for a mid to top tier graphics card. eg gtx 670,680, 660ti, hd 7970,7950,7870?

My soldering iron awaits!

Brilliant work Mux

_/-\o_

Door Tweakers user mux, vrijdag 04 januari 2013 17:15

I'm thinking of doing something like that for my next endeavor, but don't expect that to be anytime soon. I only do 1 of these large projects a year.

Door OB-1, woensdag 09 januari 2013 05:14

G'Day Mux,

You have produced amazing work (yet again) that really kicks ass, awesome stuff!

Dude I am a Senior Industrial Designer (Product Design/UI/UX) that would love to work/collaboarte with you (remotely I live in Australia). Drop me a line.

Kind Regards,
OB-1

Door Ashely, dinsdag 15 januari 2013 05:35

I'm way out of my depth now -- but sounds really interesting, anyway :)
Well, if you do ever manage to make an ultra-efficient Nvidia or AMD card... lemme know and I'm sure I'll write it up!

Door Tweakers user Chip., zondag 10 februari 2013 12:55

Ben je bekend met het compsoc platform? http://compsoc.eu/

Door cmatt85, dinsdag 02 april 2013 21:38

Where can I learn more about your cpu voltmods?

Door PvSwie, maandag 13 mei 2013 22:05

Hi,
I have had running a low power server for 20 years by now (well in fact they got replaced every so many years). I never reached you power numbers and am highly impressed with them. I am now up for building something like fluffy2 and have a few questions:
- I intend it to run ESXi. Yes I know there is a separate section for that. Do you have any experience with, can you give an estmate of what Fluffy2 would consume assuming there would be minimal CPU load and power consumption of data storage (mechanical disks) is excluded. The CPUs of the DQ77KB all have graphics build in. Is it possible to completely disable this from BIOS. How would that impact power consumption?
- What is the power usage impact of using (1) different brands of memory? (2) different speeds of memory? both idle and under load?
- Can you give a few suggestions for highly efficient 150 or 100 W 19V power bricks. How (in-)efficient are these power bricks? You write to have done mods for the power supply. Where can I find these?
- Where can I find details about your DIY UPS?

Door Tweakers user CmdJohnson, zondag 11 augustus 2013 11:59

Good read. Excellent English. Keep up the good work!

Door Tweakers user Alexxxxxxxxxx, maandag 09 september 2013 23:11

How expensive will this be alltogether when buying / building something like this for myself?

Reageren is niet meer mogelijk