Alltagstaugliche Desktop-Systeme mit 10W Idle-Verbrauch (inkl. Llano FM1 Beispiel)

[kaskode]

DEMUX, thanks for the prompt and exhaustive answer.

As for the second part, that is exactly the information I was looking for. I will go looking at lostcircuits.

As for the first part, I am still not sure I am getting your point (but then, I am not an EE major). While there might a theoretical advantage to have the intermediate conversion voltage at the square root of the two, I am pretty confident that this effect will be swamped by the actual implementation (topology, controller and quality of switches and chokes). Again, just pointing out circumstantial evidence: why would Intel for this particular motherboard as well as Dell, IBM (now Lenovo), HP and most Japanese laptop makers choose an external power supply voltage of about 20 V when they are based in and have their main markets in countries with 100 to 120 V mains? According to your formula, the optimum conversion point between 180 V rectified and filtered mains and something like 1.3 V operating voltage would be closer to 12 V.

So what we should be doing is exactly what has been done for AMD FM1 and Bobcat/Zacate boards:
- choose those that have both good idle power as well as full load power to identify those with efficient on board power conversion
- among those, chose those with the BIOS most amenable to undervolting CPU, IGP/NB and residual chipset

[demultiplexer]

DEMUX, thanks for the prompt and exhaustive answer.

As for the second part, that is exactly the information I was looking for. I will go looking at lostcircuits.

As for the first part, I am still not sure I am getting your point (but then, I am not an EE major). While there might a theoretical advantage to have the intermediate conversion voltage at the square root of the two, I am pretty confident that this effect will be swamped by the actual implementation (topology, controller and quality of switches and chokes). Again, just pointing out circumstantial evidence: why would Intel for this particular motherboard as well as Dell, IBM (now Lenovo), HP and most Japanese laptop makers choose an external power supply voltage of about 20 V when they are based in and have their main markets in countries with 100 to 120 V mains? According to your formula, the optimum conversion point between 180 V rectified and filtered mains and something like 1.3 V operating voltage would be closer to 12 V.

Of course, the above effects I described ony make sense with all else being equal. First of all: the formula still stands because universal input power adapters have either a PFC circuit that doubles the input voltage (to get back to 320VDC) or use other techniques that basically boil down to the same idea (primary voltage feedforward, etc.). But sure, if you use a simpler power adapter (cheapo power adapters without PFC, simple forward topology inside), going from 115V to CPU voltage would be better suited to a 12V intermediate step. But a second effect that I forgot to mention in the last post is that there is a slight upward bias in the ideal intermediate voltage that becomes more pronounced as current goes up. Simply because at 19V, the current required for a certain power is lower, the joule losses in anything from the mains isolation transformer to the power line switching and VRM MOSFETs on the motherboard are reduced. The general rule of thumb is: the larger the distance that needs to be crossed, the higher this bias towards higher voltages. That is why we use true high voltage to go from power plants to neighborhoods, 230V to go from the center of the neighborhood to your house, 19V to go from the mains socket to the motherboard and 1V to go from the VRM to the CPU itself.

With respect to the 'all else being equal' condition: of course here are also things I inspected before buying the board. I looked for a VRM configuration with the lowest possible output current design. DQ77KB and DH61AG (very similar boards) both have a 65W TDP (and a design output current of 99A), which is very low indeed. Also, although this is extremely common nowadays, I was looking for a board with intel SVID mode control and intelligent phase switching. In the past, dc/dc controllers for CPUs needed to respond to sudden changes in required current with an analog control loop which meant that for transient response purposes those converters always were a bit conservatively tuned. They had to anticipate that suddenly the CPU would draw much more current, so only with very tight (relatively inefficient) control loops could you really turn off all VRM phases but one. Nowadays, through SVID the processor can say in advance that it will need e.g. 50A in the next couple µs, so the converter can turn on and off phases a bit easier and can use a slower/more efficient control loop. Again, there is much more to it than this but this should at least give some insight as to why I was confident with the power distribution in DQ77KB in advance. It's more than just the 19V thing (I also mentioned that that was only one aspect in the last post, I had to finish up that post quickly because somebody was calling me on skype).

So what we should be doing is exactly what has been done for AMD FM1 and Bobcat/Zacate boards:
- choose those that have both good idle power as well as full load power to identify those with efficient on board power conversion
- among those, chose those with the BIOS most amenable to undervolting CPU, IGP/NB and residual chipset

Certainly, but just by using at experimental results (and restricting yourself to softmods/bios-based undervolting) you can easily miss things. In order to really get to the absolute lowest power consumption (and in order to be able to reliably predict power consumption in advance) you can use a lot of EE principles. I'm sure I am not the only one applying hardcore power conversion theory to computers, in the overclocking community there are a couple of EE M.Sc./Ph.D.'s as well.

[kaskode]

Would you happen to know whether the DH61AG and DH61DL both have this SVID control? The AG has the 19 V input, wheraes the DL has ATX. Over at Meisterkuehler.de, someone reached 15 W Win7 idle out of the box with an AG, a G530 and some NTS 120 W external supply, whereas another person reached 14 - 15 W idle with the DL board, G620T, Lenovo 72 W PS, WI-120 Pico and Linux.

Also, are you aware of any other boards (preferably with PCI) that use SVID? edit: it seems MSI have a BIOS/EFI option of choosing between APS and SVID. One guy with an MSI H67 based board droppe dropped idle powe from 23.6 to 17.7 W by switching from APS to SVID settings.

I think you also mentioned that you were going to use the LVDS interface. Is it pretty straightforward by now to use this interface to connect a laptop or desktop panel? I could not find any standardized cables for this, and from replacing a laptop screen I know that the only thing that is somewhat standardized it the LVDS connector on the panel.

And yet another question: I presume the wireless keyboard uses a USB receiver? Will removing that further reduce power consumption?

[demultiplexer]

All Ivy Bridge motherboards have to support intel SVID, but not all boards support a straight implementation. This is why there is such a large difference between APS (active phase switching) and SVID mode on some boards. Intel SVID is how the processor itself communicates with the VRMs. In the past, Intel used parallel VID which was an 8-bit binary number representing the voltage that the processor requested. 0b00000000 was 0.5V and 0b11111111 was - I think - 1.625V. This was very easy to hack because obviously you could just interrupt these lines and do something else with them. Nowadays intel uses a subset of i2c, which they call serial VID (SVID), to communicate this information - and more. That 'and more' is what I talked about in the last post. There is no question that all boards that support ivy bridge also support SVID because the old style VID doesn't exist anymore, but the processor still needs to be able to communicate voltages to the VRMs. The difference is whether the board directly lets the cpu communicate with the VRM (SVID mode) or whether the super I/O chip or some other chip is used as an intermediate, intercepting SVID commands and possibly changing them along the way.

Most motherboards that do this intercepting do this in a silly way. They only look at the VID messages (so that you can for instance offset processor voltage by -0.1V), and throw away all other commands. MSI is one of those manufacturers that does that, at least on my MSI H67MA-E35 and H61MA-E33 boards.

Intel implements SVID flawlessly, obviously because they're the ones who thought it up (and would like everyone to use it). All Intel boards do proper SVID mode operation, even the boards that have under/overvolting options. I did hear of the high-end boards (X78, etc.) having some quirks, but these were BIOS errors and were fixed after a while. Intel in general has very efficient boards, they make power efficiency their core business (because a lot of their profits come from laptop components). Their boards are usually a good choice for low power consumption, but they are not up to the same level of component quality as for instance mid/high-range Gigabyte boards. Even the more expensive boards like DQ77KB IMO don't give enough bang for buck.

The LVDS interface is not at all straightforward or easy to do. It's something to be left up to system integrators, not amateurs (unfortunately). To make it work I had to design this little PCB: link. This will not change in the near future, because the industry is now slowly transitioning from LVDS (and TMDS) to embedded displayport (eDP). There is no standardized eDP connector yet. There is also no standardized backlight inverter solution, even DH61AG/DQ77KB have multiple nonstandard connectors to perform various backlight functions.

My keyboard/mouse combo uses one of those unifying receivers. It adds only 0.1W to the power consumption of the PC. There are reports of motherboards that suddenly use much more power (3-6W more) when any USB device is plugged in w.r.t. no USB devices at all. This seems to be some kind of BIOS/EFI bug that has to do with the processor not being able to go into C6 mode. I have not encountered this problem myself yet.

[kaskode]

Demux, thanks again for the quick and comprehensive answer. Please excuse my asking so
many questions - it seems I have some catching up to do after three years' absence from undervolting desktop hardware.

Surely when you write ivy brigde, you also mean include sandy as they share the interface?

So let me try to summarize what I think I have understood:
- all 1155 boards interface to SVID
- there seems to be no easy way to have a frequency dependent undervolting as could be implemented for So939/AM2 Athlons and first and second generation cores
- some boards will intercept the SVID and add a user or BIOS defined offset but this offset is always the same
- there are some non-voltage SVID commands - what are these?
- most or all boards should also have a mode in which they execute the SVID verbatim

From this, one should try to find non-Intel boards that have a "pure" SVID mode, decent power converters and not too much unneccesary peripherial chips. This kind of board should then be as efficient or maybe even better than the DQ77KB?

Your LVDS adapter looks like it is just an electrical adapter, and the only extra components are a switching regulator - correct?

The USB thing was a issue with most Athlon and i915 boards. As this thread seems to imply, even some 1155 boards suffer from tis problem. It is reassuring to know that there are also good implementations.

[fp483]

Kurzes Update zum asrock q77m:

Ich habe vom ASROCk support ein nicht offizielles BIOS erhalten, mit dem sich die Vcore nun in 50mV schritten von -300mV bis +300mV einstellen läßt. Der Stromverbrauch liegt im Idle dennoch etwa 0.5W (bei 20-20.5 W) höher bei gleichen Einstellungen, allerdings könnte das auch vom distributions her kommen.

Ein Test mit dem E9-400W kann ich leider erst machen, wenn sich die Möglichkeit für den Umbau ergibt. Meine Holde hat sich leider derart an den videorekoder/htpc/server gewöhnt, dass ein Ausfall nur schwer tolleriert wird;-)

gruß dsred

[Dr.Faust]

Hier mal ein sehr interessantes Ergebniss mit dem Intel DH77DF:
- Intel DH77DF (Dry Fork) ITX motherboard - Page 3 - [H]ard|Forum

Wenn man mal mit den Ergebnissen rechnet (wobei fraglich ist, ob das legitim ist):
- Video Ausgang an und beide Lüfter aus (in der Kombi nicht getestet, auch keine Aussagen ob Maus/Tastaturangeschlossen)
-> 10W

[kaskode]

und noch ein Link (Unterlink aus Fausts Link):
SPCR • View topic - DH61DL vs. DH61AG power test ... and G620 vs G620T

Demnach:
DH61DL mit Meanwell 12 V - Supply und normaler 12 V Pico-PSU, G620 12 W idle, mit G620T eher 1 - 2 W mehr, unter Last aber 5 W weniger
DH61AG mit einem flachen Dell-Netzteil (müßte bereits Klasse V sein, also auch effizient) 19 W idle, auch unter Last eher 6 W mehr

Das schlechte Abschneiden des AG wird auf das Dell-Netzteil geschoben. Ich glaube, ich komme an so eins ran, werde mal den Wirkungsgrad nachmessen.

Interessant auch, daß der G620T im Idle eher schlechter ist, unter Last zwar besser, normiert auf die MHz aber schlechter ist. Kann natürlich im konkreten Fall an den Exemplaren liegen, aber allmählich häufen sich die Ergebnisse sehr guter Idle-Performance mit 65 W - Intel - CPUs.


Edit: das Meanwell ist laut Datenblatt Effizienzklasse IV (nach der Einteilung also schlechter als das Dell) und hat bei voller Last und 230 V Eingang 88% Wirkungsgrad. Das sagt natürlich nichts über die Effizienz bei 12 W aus...

[demultiplexer]

Demux, thanks again for the quick and comprehensive answer. Please excuse my asking so
many questions - it seems I have some catching up to do after three years' absence from undervolting desktop hardware.

I doubt you will find this information anywhere else without doing some original research yourself.

Surely when you write ivy brigde, you also mean include sandy as they share the interface?

As I understand it, both VRD11.1 and VRD12 specifications are supposed to be compatible with sandy bridge (VRD, or voltage regulator down, being intel's specification for their VRM interface), which means that it may be possible to find s1155 sandy bridge motherboards with an old-style parallel VID VRM. Not sure about this, and I have never seen such a board in the wild.

So let me try to summarize what I think I have understood:
- all 1155 boards interface to SVID
- there seems to be no easy way to have a frequency dependent undervolting as could be implemented for So939/AM2 Athlons and first and second generation cores
- some boards will intercept the SVID and add a user or BIOS defined offset but this offset is always the same
- there are some non-voltage SVID commands - what are these?
- most or all boards should also have a mode in which they execute the SVID verbatim

yes, but noting that the second point is very easy to circumvent, it just hasn't been done before AFAIK. It would require cutting traces and soldering very fine wires on the board. The non-voltage SVID commands are things like handshaking information, the VRM telling the CPU what features it supports, etc. I have a datasheet for NCP6151 that details some of the commands, maybe tonight I can upload it somewhere. Usually this information is under NDA, and I have an intel developer account so that is where I get that information. Unfortunately stuff you download there has watermarks, so if I spread it I might get banned.

From this, one should try to find non-Intel boards that have a "pure" SVID mode, decent power converters and not too much unneccesary peripherial chips. This kind of board should then be as efficient or maybe even better than the DQ77KB?

Basically, yes. I would be very surprised if a board really exceeds the combined efficiency of DQ77KB+a decent 19v adapter, though. Most alternatives will require a picopsu (which is a bit less efficient than purpose-built on-board conversion, and runs at a slightly less favourable voltage) and have not been built with energy efficiency in mind. DQ77KB is one of a couple of thin mini itx boards, which have efficiency as a high design priority.

Your LVDS adapter looks like it is just an electrical adapter, and the only extra components are a switching regulator - correct?

Yep.

The USB thing was a issue with most Athlon and i915 boards. As this thread seems to imply, even some 1155 boards suffer from tis problem. It is reassuring to know that there are also good implementations.

I don't think this issue is even intel specific, I've heard of plenty of AMD motherboards doing the same thing. Something as simple as those old-style NEC USB 3.0 controllers not being configured for auto power down makes a ~1.5-2W difference already. It's like DTS, HDMI CEC and CIR support - motherboard manufacturers always seem to screw those little details up.

edit: about those G620/G620T results: the guy reports very large differences in power consumption in 'controlled' tests (i.e. saying it's 12-14W, most of the time 12W). That tells me that stuff is going on in the background and the test is basically not conducted right. I don't think you should read any valuable information into that. Same goes for efficiency classes; even the best currently enforced efficiency class (class V) is crap for anything but the lowest rated power supplies (i.e. 65W and less), because all 80W+ power adapters I have tested, from crappy to great, manage to get 87-93% efficiency at typical power levels. Even stuff from like 2001. There is no reason to look at efficiency classes at all, only at real measurement data.

[celemine1Gig]

...
Demnach:
DH61DL mit Meanwell 12 V - Supply und normaler 12 V Pico-PSU, G620 12 W idle, mit G620T eher 1 - 2 W mehr, unter Last aber 5 W weniger
...

Kann ich so bestätigen. Habe mit genau der Konfiguration (DH61DL + 620T + 8GB RAM + SSD + Pico-150XT + Morex 80W) im Idle ca. 12W gemessen. Niedrigste Werte um die 11.x W, im Schnitt wohl eher 12-12.5W. Gemessen mit einem KD302.

[Mars Warrior]

Ich habe jezt zwei mainboards ausprobiert: Intel DH77DF und Asus P8H77-I met folgenden componenten:
- Altes WD 500GB (9W idle!)
- 8GB Sniper 1.25V ram
- WD 2.5TB & WD3.0TB festplatten (~~0.9W standby)
- Intel 3570K CPU
- CPU-Lüfter (0.7W)
- Be Quiet F1 Efficient 350W Gold PSU
- C1/C3/C6 eingeschaltet

Asus P8H77-I:
- 28.6W (Monitor aus, kein USB, 2.5TB und 3.0TB standby)
- 30.0W (Monitor, kein USB)
- 37.4W (Monitor, Dell USB Tastatur & Logitech Unifying Receiver. Receiver nimmt 6W auf dieses mainbord!!!!!)

Dabei macht dieses mainboard ganz viel lärm
RAM laüft auf 1.23V ohne probleme.

Intel DH77DF:
- 23.4W (Monitor aus, kein USB, 2.5TB und 3.0TB standby)
- 25.0W (Monitor, kein USB)
- 27.0W (Monitor, Dell USB Tastatur & Logitech Unifying Receiver. Also kein einziges problem mit dem receiver auf dieses mainbord!!!!!)

Deises mainboard ist sehr, sehr leise.
RAM laüft auf 1.55V (1.4V im BIOS). Weniger geht nicht, mainboard POST nicht mehr unter 1.4V...

Also mit nichts angeslossen ist das unterschied circa 5W. Und mit USB bis zu 10W...

Ohne die drei festplatten, aber mit SSD, ohne BIOS RAM bug und Pico wird bestimmt die 10W möglich denke Ich.

Messgerät: Energy Logger 4000
OS: WHS2011 mit Stablebit DrivePool, SABnzbd, ORB Caster & DLNA, Gallery Server Pro und AirVideo installiert.

[moeller]

Llano- PC mit wenig Idle Verbrauch.
==========================
Ich möchte zwei Fragen in die Runde werfen:
1) Hat jemand Erfahrungen mit folgendem ITX - Board: Asrock A75M-ITX. Wie sind Einstellmöglichkeiten gegenüber der Referenz Gigabyte GA-A75N-USB3 ? Stromverbrauch im Idle ?
2) Llano: Ich möchte den A8-3870K untertakten und undervolten damit die "Thermal Design Power" sinkt (Kühlkörper). Macht das Sinn oder liege ich da falsch ?

[FrauKrauter]

2) Llano: Ich möchte den A8-3870K untertakten und undervolten damit die "Thermal Design Power" sinkt (Kühlkörper). Macht das Sinn oder liege ich da falsch ?

untervolten ist gut
untertakten ist schwachsinn
dann kannst du dir auch gleich nen kleineren prozessor kaufen

vollastverbrauch ist bei normalgebrauch (90% idle) nebensächlich
außer deine kühlung ist unterdimensioniert
oder du lässt das teil 24/7 irgendwelche aminosäuren falten

p.s.
willkommen im forum

[moeller]

Das wird ein Surf-Rechner, daher zählt nur der Idle-Verbrauch.
Hin und wieder spiele ich auch Videos auf dem Rechner. Aber das ist selten.

Untertakten will ich, weil ich dann weiter untervolten kann -- oder ?
Die Kühlung ist definitiv unterdimensioniert. ITX - Gehäuse => flacher Kühler.

[FrauKrauter]

dann kauf dir einen kleineren prozessor
ggf. reicht auch ein doppelkern bei den anforderungen

für so ein eingeengtes gehäuse ist das sowieso besser
und er kostet die hälfte

[moeller]

Vielen Dank für den Rat, ich werde noch mal schauen.
Zum halben Preis gibt es den A4-3400, der nach che new (Eröffnungsbeitrag dieses Threads)
etwas mehr idle-Leistung braucht als ein A6-3670.
Aber das Stromsparen darf nicht zu teuer werden

[Xer0]

Wow, nicht schlecht. hätt nicht gedacht das AMD die sonst so super effizienten Ivy's schlagen kann, im Ilde zumindest.
wobei schwache Last bei mir auch das längste Nutzungsszenario darstellen würde...

Wie schlägt sich der A8 Liano gegen einen Phenom X4 B55? (unlocked Athlon X2 555)
Hab so einen hier stehen inkl GPA-890-UD3H Board und der idle't nicht unter 30W!

[FrauKrauter]

llano ist vom prinzip her besser
35nm, northbridge und grafik auf einem die

praktisch hängts sehr stark von der umgebung ab
das system in meiner sig kommt mit allen klimmzugen auch auf 30W idle

[thom_cat]

gut, dass ich das hier wieder lese.

habe das a300 und das e9 400 da, werde mal schauen wie die beiden sich zueinander verhalten.
denke auch nicht, dass das e9 da noch was rausholen kann.
das a300 hat bei mir im idle sogar das huntkey jumper gold geschlagen.

Ich aber auch...
Fakt ist, das E9 schlägt bei meiner Config das LC7300 und ist unhörbar.
PC-Experience Reviews : | be quiet! Straight Power E9 400Watt Netzteil im Test

Das E9 kostet aber auch fast das doppelte. Ich hab das E9 500W in meinem Zockrechner und es ist wirklich unhörbar, aber wenn es wirklich bei 10-20W Verbrauch besser sein sollte als das A300 (was ich bezweifel), dann muss ich den Rechner 10 Jahre laufen lasse, bis ich diese Mehrkosten durch den hohen Kaufpreis wieder drinnen habe.
Unter Last glaube ich das sofort, aber im absoluten Idle Bereich, kommt nicht viel an dem A300 vorbei - zumindest im normalen ATX Bereich.
Was übrigens auch damals in 2/2012 von der c't bestätigt worden ist.

Ob bei einem Verbrauch von max. 20W ein Netzteil schliesslich 70, 80 oder sogar 90% Effizienz hat, macht kaum einen Unterschied. 70% zu 80% sind dann vielleicht 1-2W, das macht vielleicht hier in Benchmarks einen Unterschied, aber in der Realität eben nicht.

so, mal ein paar quotes gesammelt damit das thema auch zusammen passt

habe meine tests jetzt endlich geschafft und hier das ergebnis:

• Cougar A300
• Huntkey Jumper 300G
• Be Quiet Straight Power E9 400 Watt

Testsystem:

Intel I3 2100
Intel BH67BL
4GB G.Skill Eco 1333
Solidata K5 64GB
Intel Postville G2 80GB
DVD Brenner


Cougar A300:

Idle: 22 Watt
Prime95: 68 Watt


Huntkey Jumper 300G:

Idle: 22 Watt
Prime95: 64 Watt


Be Quiet Straight Power E9 400:

Idle: 19 Watt
Prime95: 62 Watt

das be quiet schlägt das cougar a300 also tatsächlich im idle betrieb!
so deutlich hätte ich das nicht erwartet... allerdings ist der aufpreis natürlich sehr hoch und mit der ersparnis nur über jahre wieder rauszuholen

[bawder]

das be quiet schlägt das cougar a300 also tatsächlich im idle betrieb!
so deutlich hätte ich das nicht erwartet... allerdings ist der aufpreis natürlich sehr hoch und mit der ersparnis nur über jahre wieder rauszuholen

ich wiederrum schon, weil das cougar wie viele viele andere, billige netzteile auf 60-75% wirkungsgrad runtergeht, unter 20% last.

edit: sieht man ja auch bei E9 vs. F1; in zukunft kauft man also keine netzteil mehr aufgrund der Watt-zahl

[thom_cat]

übertreib es mal nicht mit deinen prozentzahlen... es bleibt unterm strich trotzdem die bessere wahl für kleine günstige systeme.

das e9 kann die ersparnis auf jahre nicht reinholen, so viel sollte auch klar sein.

[bawder]

übertreib es mal nicht mit deinen prozentzahlen... es bleibt unterm strich trotzdem die bessere wahl für kleine günstige systeme.

das e9 kann die ersparnis auf jahre nicht reinholen, so viel sollte auch klar sein.

was heißt übertreiben, das is einfach so. das cougar würde auf ca. 69% kommen wenn ich das E9 bei 80% wirkungsgrad ansetze. (80% idle für 19W auf dem strommessgerät)

edit: zu den kosten, das sind 24€ differenz; im jahr kostet ein Watt maximal 2,19€, dann wären die 24 € nach 3,5 jahren erreicht. läuft der PC einen halben tag lange, dann nach 7 jahren.

[thom_cat]

wie gesagt, das sind schon richtig lange laufzeiten die man da haben muss bevor es sich wirklich lohnt.

[FrauKrauter]

wer betreibt seine hardware 7 jahre lang? in 3 jahren gibts wieder was noch effizienteres
und wenns wirklich eine 24/7 maschine ist
ists meist ein kleiner homeserver
und da punkten immer noch picos

energieeffizienz gut und schön
aber wenns nicht als sport betrieben wird ist das kosten/nutzen verhältnis ausschlaggebend

[bawder]

wer betreibt seine hardware 7 jahre lang? in 3 jahren gibts wieder was noch effizienteres

bei mir und den meisten home-usern werden die netzeile richtig alt, da lohnt es sich meiner meinung nach.

---------- Post added at 13:09 ---------- Previous post was at 13:09 ----------

und da punkten immer noch picos

NUR wenn sie ein sehr gutes netzteil haben.