Google suggest new 12V-only PSU standard
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee, Devonavar
-
- Posts: 81
- Joined: Sun Oct 09, 2005 7:53 pm
Google suggest new 12V-only PSU standard
Googleblog (with link to white paper):
http://googleblog.blogspot.com/2006/09/ ... uting.html
NYT reporting:
http://www.nytimes.com/2006/09/26/techn ... ref=slogin
The idea is motherboards which take just 12V DC, and do the other required DC conversions on board. This simplifies the PSU, allowing it to be cheaper and more efficient, and also allows datacenteres to centralize their AC to 12V DC conversions, and avoids lots of conversion losses with UPSs.
http://googleblog.blogspot.com/2006/09/ ... uting.html
NYT reporting:
http://www.nytimes.com/2006/09/26/techn ... ref=slogin
The idea is motherboards which take just 12V DC, and do the other required DC conversions on board. This simplifies the PSU, allowing it to be cheaper and more efficient, and also allows datacenteres to centralize their AC to 12V DC conversions, and avoids lots of conversion losses with UPSs.
So that the motherboard becomes more complex with layout problems, more expensive, less efficient and hotter.he idea is motherboards which take just 12V DC, and do the other required DC conversions on board. This simplifies the PSU, allowing it to be cheaper and more efficient, and also allows datacenteres to centralize their AC to 12V DC conversions, and avoids lots of conversion losses with UPSs.
The latter is a valid arguement but only in the server market. I cannot see where this would provide much consumer benefit though in a datacenter full of rackmounted servers a specialized mobo/psu setup would be of benefit.
-
- *Lifetime Patron*
- Posts: 618
- Joined: Mon May 23, 2005 7:05 am
- Location: State College, PA
the question would be whether the AC to 12V (PSU) and then DC to DC conversion (MB) would be any more efficient overall than AC to 3.3, 5 +12V DC at the mo'. Given how no-one really knows how efficient MB DC to DC conversion is even for just the CPU power circuit (compared with the transparancy/ testing of PSU efficiencies) and that changing a MB is far more hassle than changing a PSU, this isn't neccessarily a great step forward.
Changing everything in a PC to work on one DC voltage, now that would help things a bunch. Doubt it'll happen for a while though.
Changing everything in a PC to work on one DC voltage, now that would help things a bunch. Doubt it'll happen for a while though.
-
- Posts: 81
- Joined: Sun Oct 09, 2005 7:53 pm
My understanding of their argument is:
1) Motherboards already have to do all this voltage changing - there is little or no extra complexity on the m/b to take only 12V.
2) With the current system, some motherboards will draw much more (e.g.) 5V than others - so the PSU has to be designed for the high 5V demand motherboards. But most motherboards will draw much less than this - so the 5V circuits are usually running at much less than full capacity. The circuit is more efficient at near full capacity than at low capacity, so this is inefficient.
I didn't figure out what they intended to do about drives and other peripherals. I'm guessing they need to be redesigned to take 12V-only also.
Anyway, they are claiming over 90% efficiency in AC to DC conversion in Google datacenters due to implementing this stuff. Follow the link in the initial post to read their whitepaper.
1) Motherboards already have to do all this voltage changing - there is little or no extra complexity on the m/b to take only 12V.
2) With the current system, some motherboards will draw much more (e.g.) 5V than others - so the PSU has to be designed for the high 5V demand motherboards. But most motherboards will draw much less than this - so the 5V circuits are usually running at much less than full capacity. The circuit is more efficient at near full capacity than at low capacity, so this is inefficient.
I didn't figure out what they intended to do about drives and other peripherals. I'm guessing they need to be redesigned to take 12V-only also.
Anyway, they are claiming over 90% efficiency in AC to DC conversion in Google datacenters due to implementing this stuff. Follow the link in the initial post to read their whitepaper.
-
- *Lifetime Patron*
- Posts: 618
- Joined: Mon May 23, 2005 7:05 am
- Location: State College, PA
sure AC- single DC conversion can easily reach 90% (I think most laptop bricks run to 85-90% efficiency) but when you add in the DC-DC efficiencies you often run to the same efficiencies as AC- multiple DC conversion - I think the PicoPSU review is a great eg. of that. At the middle of it's operating range it's only 1 or 2% more efficient overall than a high quality single PSU, though the efficiency curve certainly appears flatter.
Other potential issues:
- cooling. Going by the PicoPSU, roughly half of the heat generated by a PSU is from the DC-DC conversion. Granted in a workstation or standard desktop, that'll result in 5-10W waste heat (easily coolable) but what about higher end rigs? There's not an awful lot of spare realestate available on MBs for elaborate cooling solutions.
- compatibility. Designing power conversion circuits for server or workstation boards at their optimal efficiency should be relatively straight forward as power loads and usages are both relatively predictable and also dictated by the system integrator. What about aftermarket boards? Will every MB be designed to handle the most powerful rigs possible with that board, therefore operating at inefficient power levels with modest rigs, or will there be multiple MBs of the same design with different power ratings?
For servers, sure I can see how it'd work, but for every day PCs it doesn't appear anywhere near so clear cut. Certainly a step in the right direction though
Other potential issues:
- cooling. Going by the PicoPSU, roughly half of the heat generated by a PSU is from the DC-DC conversion. Granted in a workstation or standard desktop, that'll result in 5-10W waste heat (easily coolable) but what about higher end rigs? There's not an awful lot of spare realestate available on MBs for elaborate cooling solutions.
- compatibility. Designing power conversion circuits for server or workstation boards at their optimal efficiency should be relatively straight forward as power loads and usages are both relatively predictable and also dictated by the system integrator. What about aftermarket boards? Will every MB be designed to handle the most powerful rigs possible with that board, therefore operating at inefficient power levels with modest rigs, or will there be multiple MBs of the same design with different power ratings?
For servers, sure I can see how it'd work, but for every day PCs it doesn't appear anywhere near so clear cut. Certainly a step in the right direction though
-
- Patron of SPCR
- Posts: 2674
- Joined: Sun Jul 04, 2004 6:07 am
- Location: Houten, The Netherlands, Europe
Only the 12V to 5V, 3.3V, etc. has to be handled by the mobo. The difference between a low and high power rig is mostly on the 12V rail.mattthemuppet wrote:What about aftermarket boards? Will every MB be designed to handle the most powerful rigs possible with that board, therefore operating at inefficient power levels with modest rigs, or will there be multiple MBs of the same design with different power ratings?
It shouldn't be too hard to modify the design of the pico PSU, so it can pass more 12V current. There you have the solution for the lack of space for new board real-estate too.
-
- *Lifetime Patron*
- Posts: 618
- Joined: Mon May 23, 2005 7:05 am
- Location: State College, PA
true,hadn't thought of that.
In my mind the limitation for the PicoPSU has been finding fanless bricks over 100W, but that seems to be improving with 200W fanless bricks coming out. AFAIK the PicoPSU passes 12V straight through so it's 12V capacity should be theoretically limitless - however I don't know it's real limit.
In my mind the limitation for the PicoPSU has been finding fanless bricks over 100W, but that seems to be improving with 200W fanless bricks coming out. AFAIK the PicoPSU passes 12V straight through so it's 12V capacity should be theoretically limitless - however I don't know it's real limit.
-
- Posts: 107
- Joined: Sat Feb 05, 2005 3:45 am
- Location: Moscow, Russia
- Contact:
1) SilverStone Zeus SST-ST75ZF and SST-ST85ZF have one AC-DC converter and two fully independent DC-DC PWM converters (STMicro L6730 PWM controllers with synchronous rectifiers and 400 kHz switching frequency). Their overall efficiency is about 85% -- good, but nothing exceptional.
2) Distributed power systems (DPS) are widely used in servers: one primary 220VAC - 48VDC PSU and multiple secondary PSUs. Nothing new. Lowering the intermediate voltage from 48 VDC to 12 VDC is a very bad idea.
2) Distributed power systems (DPS) are widely used in servers: one primary 220VAC - 48VDC PSU and multiple secondary PSUs. Nothing new. Lowering the intermediate voltage from 48 VDC to 12 VDC is a very bad idea.
reminds me of something my uncle has told me some years ago, he said in coming years less and less computer equipment would need the other rails and almost all load would move onto the 12V, well that certainly came true, and he went on to explain the best new standard for computer power supplies would be to have only a -12V rail. i forget the reason for the negative, but it was convincing at the time..
You'd also need hard drives to work on 12V only.
I was thinking about the PicoPSU for a new build I am doing, but I don't think it will have enough juice for the drives: I think the system will need more than 90W. The real problem here is not the motherboard but the difficulty in finding a high current AC-12VDC converter (apart from a full-blown PSU, which defeats the object).
But a high current 12V source is going to need a very thick cable to keep down resistive losses, bearing in mind that an external power brick probably also has a longer cable anyhow. For example 240W at 12V would be 20 amps of current: that would need a seriously thick cable or probably a whole bunch of black and yellow. Remember that cable power losses follow the I^2R law, so for example a 0.01 ohm cable would dissipate 4W of power. Any cable damage and the thing will probably melt or catch fire. Like Oleg Artamonov says, 48V is a much better idea. (Well this is maybe a bit too alarmist. My 200W music system could be pushing 50A through the speaker cables, if the speakers are 4W. I guess you just need good fat loudspeaker-style cables for a 12V rail system.)
I was thinking about the PicoPSU for a new build I am doing, but I don't think it will have enough juice for the drives: I think the system will need more than 90W. The real problem here is not the motherboard but the difficulty in finding a high current AC-12VDC converter (apart from a full-blown PSU, which defeats the object).
But a high current 12V source is going to need a very thick cable to keep down resistive losses, bearing in mind that an external power brick probably also has a longer cable anyhow. For example 240W at 12V would be 20 amps of current: that would need a seriously thick cable or probably a whole bunch of black and yellow. Remember that cable power losses follow the I^2R law, so for example a 0.01 ohm cable would dissipate 4W of power. Any cable damage and the thing will probably melt or catch fire. Like Oleg Artamonov says, 48V is a much better idea. (Well this is maybe a bit too alarmist. My 200W music system could be pushing 50A through the speaker cables, if the speakers are 4W. I guess you just need good fat loudspeaker-style cables for a 12V rail system.)
It seems to me Google's proposal would scatter the PSU's function all about the PC, requiring hard drive manufacturers to either switch over to 12V logic and motors, or place a DC-DC converter on the hard drive. Motherboards would have to generate 5V for USB, etc.
Instead, have the next ATX standard specify a small amount of power (<4A?) on the deprecated rails. That way you have support for most legacy devices, and the efficiency and cost savings of being able to do all your AC-DC conversions to 12V. Add a few DC-DC converters inside the PSU, and you have something which functions the same as a current-gen PSU, but may be more efficient.
Of course, that would leave PSU makers free to decide whether they want to build a seperate circuit for AC-5VDC, or take the 12VDC-5VDC approach.
Instead, have the next ATX standard specify a small amount of power (<4A?) on the deprecated rails. That way you have support for most legacy devices, and the efficiency and cost savings of being able to do all your AC-DC conversions to 12V. Add a few DC-DC converters inside the PSU, and you have something which functions the same as a current-gen PSU, but may be more efficient.
Of course, that would leave PSU makers free to decide whether they want to build a seperate circuit for AC-5VDC, or take the 12VDC-5VDC approach.
-
- Posts: 107
- Joined: Sat Feb 05, 2005 3:45 am
- Location: Moscow, Russia
- Contact:
No, it wouldn't. Independent DC-DC converters are not more efficient than group stabilization (used in most of the modern PSUs) or mag-amps (used in PSUs with independent voltage stabilization such as Antec, Tt ToughPower or FSP Zen; it's independent only from user's viewpoint, really mag-amp is a part of main PWM converter and can't work standalone).Brian wrote:Add a few DC-DC converters inside the PSU, and you have something which functions the same as a current-gen PSU, but may be more efficient
There is no separate AC-5VDC circuit in PSU. And there is no 12VDC-5VDC converters. PC PSU schematic is somewhat different... If you need technical information, see here. If you need detailed explanation and design examples, see here, here and here.Of course, that would leave PSU makers free to decide whether they want to build a seperate circuit for AC-5VDC, or take the 12VDC-5VDC approach
But don't read google papers, it's a bullshit.