ECN Electrical Forum - Discussion Forums for Electricians, Inspectors and Related Professionals

ECN Shout Chat
Top Posters(30 Days)
Admin 19
Recent Posts
fuse rejectors
by NORCAL. 03/29/17 01:44 AM
Old decora style outlets
by NORCAL. 03/29/17 01:41 AM
NFPA stuff
by gfretwell. 03/28/17 08:08 PM
Anyone hiring inspectors?
by HotLine1. 03/27/17 08:03 AM
ESA Arc flash course
by TheShockDoctors. 03/24/17 10:15 AM
New in the Gallery:
SE cable question
Popular Topics(Views)
231,742 Are you busy
166,641 Re: Forum
160,791 Need opinion
Who's Online Now
0 registered members (), 76 guests, and 12 spiders.
Key: Admin, Global Mod, Mod
Previous Thread
Next Thread
Print Thread
Rate This Thread
#183564 - 01/14/09 07:15 PM DC power for a data center  
pdh  Offline
Joined: Jan 2005
Posts: 354
I just thought this might interest some of you. This article is another in the continuing saga of making the data center more efficient and/or less costly (competing goals, all too often).

You can also see comments posted on Slashdot here:

The concern I have with distributing DC power within a data center is that the PSU components of a computer are still needed. 12VDC-only computer board can be made (this was Google's plan) and won't need a separate PSU in the computer box. My idea would be to run 480/277 or 400/230 (depending on country of operation) directly to a power converter mounted in a 2U to 4U box in the middle of each rack cabinet. From there, 12VDC would go to each blade module in the rack. That power converter may even be combined with a short term battery backup (A UPS without the DC to AC conversion) to ride through short term interrupts in the internal distribution like 2-minute recabling.

I also expect 400VDC to be more expensive to protect than 480VAC, in part, due to the cost of the higher interruption capability needed for breakers.

Last edited by pdh; 01/14/09 07:16 PM. Reason: typo

Work Gear for Electricians and the Trades

#183580 - 01/14/09 10:13 PM Re: DC power for a data center [Re: pdh]  
gfretwell  Offline

Joined: Jul 2004
Posts: 9,060
This just doesn't make much sense to me. How is it more efficient to convert 400vdc to 12 and 5 than to convert 400vac? They both will involve a switching power supply.
In real life the "computer" itself runs on 1.5 to 3v anyway.
The disk drives are still 5 and 12. When I was doing physical planning the RAMAC racks were the most "power dense" box in the room. They are basically a rack, stuffed with 3.5" hard drives.

Greg Fretwell

#183590 - 01/15/09 01:59 PM Re: DC power for a data center [Re: gfretwell]  
pdh  Offline
Joined: Jan 2005
Posts: 354
In theory, the PSU in the computer only needs to do half the work when converting 400VDC to 12VDC/5VDC/3.3VDC (lower voltages are converted on the mainboard these days). the SMPSUs used now step the voltage up so the AC to DC conversion is done around 340V. Then the 340VDC is stepped down.

Still, by distributing DC to the computers, the AC to DC conversion has to be done somewhere. You still have all those conversions to do. So unless some particular kind of conversion can be done more efficiently at some particular voltage, it may not matter.

And this doesn't eliminate the PSU in the computer. It just makes it smaller. You're not going to feed that 400VDC straight to the mainboard and do the 400 to 12 step there. My idea completely eliminates the PSU box in each computer, allowing more other parts to be put into the same space ... in exchange for a center whole-rack-serving SMPSU.

So is 400VDC any better than 400VAC or 480VAC? I really don't see that. The conversion near the computer has less to do, but the conversion still has to be done, somewhere. At best, that would be a minimal efficiency increase. And I see a lot of new issues with such a hefty DC running around the place. The advantage of using 400VAC or 480VAC (make that a full ranging 380VAC to 480VAC) is that it allows the use of industrial-commodity distribution components, and is safer for non-electrician computer personnel to be around (pulling plugs and stuff).

How much are those 400VDC breakers going to cost compared to a 480VAC breaker?

#183592 - 01/15/09 04:30 PM Re: DC power for a data center [Re: pdh]  
gfretwell  Offline

Joined: Jul 2004
Posts: 9,060
You are adding complexity with this scheme, not reducing it. Multiple DC-DC conversions can't be as efficient as one AC-DC conversion and, as you pointed out, you still have the AC-DC conversion. The DC-DC converter has an AC intermediary anyway.
It was also pointed out up thread, interrupting big DC loads is a lot harder than AC since there is no zero crossing to put the fire out.
I predict this is a fad that will never really catch on.

Greg Fretwell

#183596 - 01/15/09 06:58 PM Re: DC power for a data center [Re: gfretwell]  
pdh  Offline
Joined: Jan 2005
Posts: 354
I hope the 400VDC through the data center doesn't catch on.

It's already being done to convert DC to DC, right on the motherboard, for all computers these days, to derive the small voltages the CPUs need. The SMPSU only provide 12V, 5V, and 3.3V. These boards could do all this with just 12VDC alone coming in. It would be a cost shift, raising the cost of the board a little bit, and reducing the need for the PSU since it only needs to supply 12VDC in this case. This is the scenario Google has asked motherboard makers to do (with an order for on the scale of a million such boards). I don't have the details of how they plan to provide 12VDC to these boards, but it has been suggested that more than one board would be supplied from a 12VDC source. What I'm trying to envision is the possible scale of converting AC to 12VDC and distributing the 12VDC from such a common supply/converter. My gut feeling is this is practical to do on the scale of one rack of about 80 or so such boards. I certainly would not run 12VDC all around the data center (way too much wire needed, way too much I2R loss, and way too much short circuit fault current). Even 48VDC on a large data center scale bothers me (100 or more rack cabinets).

So my gut feeling is that a good solution is to run 480/277 or 416/240 around the data center, feeding AC to each rack, and converting that AC to a single DC voltage that each board can take and do the remaining conversions as it needs. I believe just one AC/DC conversion per rack can be done. But if not, then maybe it can be done on 1/2 rack or 1/3 rack basis.

It gets the SMPSU box out of the computer, which is a definite advantage.

Using 416/240 would have a slight advantage that special computers and other equipment that still takes AC directly can be suppled with the 240 L-N right from that. Routers and switches are likely limited to that (though a lot of them do have 48VDC options). It would also be a common design that can be done worldwide for companies (like Google) that put in data centers in many countries.

I do know the military makes a lot of use of 416/240 and this is apparently a choice based on wide deployment compatibility issues. Look for large military surplus gensets and you find many, if not most, are made for, or are currently wired for, 416/240 or thereabout.

416/240 would have the slight disadvantage of somewhat greater I2R losses compared to 480/277, and a special order from the utility in North America. But even with 480/277, those computers can still be supplied from AC stepped down from the 480/277 via a transformer or three, either 240 L-N from 416/240, or 240 L-L from 3 sets of 120/240, or even 208 L-L if you must. I'd light the data center with fluorescent at 277 or 240, and thus limit the total amount of 120 I need to have, minimizing the size of that extra stepdown transformer. Obviously, if there are office spaces around, you need a bunch of 120 there (because you can't be having ordinary people plugging computers and other stuff into 240 in this country).

#183619 - 01/16/09 03:27 PM Re: DC power for a data center [Re: pdh]  
SteveFehr  Offline
Joined: Mar 2005
Posts: 1,198
Chesapeake, VA
A 12V bus would be massive, and wildly impractical for any size of data center. I don't see much advantage of a DC bus over an AC bus at the power supply, but one thing it DOES do is eliminate the pointless DC-AC of the UPS inverter which is immediately just stepped back AC-DC. This voltage unfortunately fluctuates between 584V float and 408V at the end of battery discharge, though, so it may not be entirely practical. But, if it's just switched power supplies, the PSUs might not really care *what* the voltage is, and be just as insensitive to DC voltage as they are to AC voltage.
Originally Posted by pdh
I do know the military makes a lot of use of 416/240 and this is apparently a choice based on wide deployment compatibility issues. Look for large military surplus gensets and you find many, if not most, are made for, or are currently wired for, 416/240 or thereabout.
416/240? Are you sure? I can't recall ever having seen one. 440V and 460V, and a slew of 400V and even some 400Hz but never 416V. 400/230V 50Hz is the standard voltage for most of the world, and just as common in the military as 208/120V 60Hz. In fact, most of the time, the generators can support any of the 3 common voltage/frequency combos with some minor tap changes to the coils (parallel or series for 480 or 208V) and a small adjustment to the controller for the specific frequency and voltage.

#183662 - 01/18/09 06:51 AM Re: DC power for a data center [Re: SteveFehr]  
pdh  Offline
Joined: Jan 2005
Posts: 354
A 12V bus for the whole data center would certainly be MASSIVE. But within ONE single rack, with separate cables to each blade frame within the rack, it may be practical. If it's borderline, then split the rack in half and do it per half-rack. Otherwise, it would be a SMPSU within each rack frame (this is already done) but that's too small to add a battery to provide the backup power.

A lot of data centers put a UPS in each rack (some as small as 2U but typically 4U to 6U in size) to power the whole rack at an AC line voltage. I don't see why that can't be the point of AC to DC conversion and just supply the DC to each blade frame (usually these are 3U to 6U themselves), eliminating the SMPSU in each frame and having more space for boards.

A lot of this is design issue, such as making blade frames without the SMPSU, and a blade system with just 12VDC.

Apparently a lot of business (maybe Google is among them) are wanting to use cheaper commodity boards instead of manufactured blades. It actually makes business sense, since 4 commodity boards can be put in, in place of 1 blade board.

How this relates to ECN is what the distribution would be within the data center itself. For THAT I really don't see enough advantage (if any at all) to go with DC (whether at 400VDC or 48VDC).

I've seen a few oddball voltages available at a few utilities a couple years ago when I surveyed a lot of their tariffs and installation requirements. I remember seeing 416/240 at two or three of them. One of them even had 240/139 available (apparently a substitute for 240 delta).

I also recall seeing 416/240 as options for pad mount transformers. That is definitely available for dry-type transformers. And it's doable with 3x 240V pole pigs as long as they don't have an internal ground on the center tap (I've been told none have that). A 480/277 meter would handle 416/240. A 480/277 panelboard would handle 416/240. The thing with 416/240 is you have 15% less capacity compared to 480/277 without increasing the amps.

I would not plug my computer into 277. But I would plug it into 240.

If SMPSU makers start making their units work up to 277 nominal, and not just 240, then that becomes a practical alternative. But that would never happen since the domestic voltages in the world are just 100 to 240.

Member Spotlight
dallas, texas, USA
Posts: 67
Joined: January 2002
Show All Member Profiles 

2017 NEC and Related
2017 NEC
Now Available!

Shout Box
Powered by UBB.threads™ PHP Forum Software 7.6.0
Page Time: 0.018s Queries: 15 (0.008s) Memory: 0.7911 MB (Peak: 0.9452 MB) Zlib enabled. Server Time: 2017-03-29 15:02:03 UTC