Stories
Slash Boxes
Comments

SoylentNews is people

posted by cmn32480 on Sunday August 07 2016, @01:08AM   Printer-friendly
from the gotta-give-back dept.

Arthur T Knackerbracket has found the following story:

Google has for the first time shared some of its data centre designs with the OpenCompute Project (OCP).

Google joined OCP back in March 2016 and, at the time, talked up its 48v racks and promised a new spec that would allow them to work with OCP servers in its own bit bars … and yours, should you choose to use OCP standards.

The Alphabet subsidiary has made good on that promise and shared a rack design in the hope they become part of the Open Rack 2.0 standard. Facebook's chipped in with some help, too.

Google says its efforts comprise a “48V power architecture with a modular, shallow-depth form factor that enables high-density deployment of OCP racks into data centers with limited space.”

The ads-and-search company says it has “extensively deployed these high-efficiency, high-availability systems since 2010.”

“We have seen significant reduction in losses and increased efficiency compared to 12V solutions,” writes technical program manager Debosmita Das and technical lead manager Mike Lau,” adding that “The improved SPUE [Server Power Usage Efficiency] with 48V has saved Google millions of dollars and kilowatt hours.”


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by Scruffy Beard 2 on Monday August 08 2016, @03:46AM

    by Scruffy Beard 2 (6030) on Monday August 08 2016, @03:46AM (#385156)

    Intel decided that servers were separate from PCs a long time ago.

    If you want ECC memory, you can't use consumer boards (Not sure if any AMD consumer boards still support ECC).

    48V PSUs for your work-station may be cheaper anyway: 4 times the voltage mean 1/4 the current: allowing thinner wires.

    Starting Score:    1  point
    Karma-Bonus Modifier   +1  

    Total Score:   2  
  • (Score: 2) by RamiK on Tuesday August 09 2016, @06:52PM

    by RamiK (1813) on Tuesday August 09 2016, @06:52PM (#385907)

    I'm fairly sure we're talking about high power supplies for 1U racks here. Regardless, this is just wrong:

    4 times the voltage mean 1/4 the current: allowing thinner wires.

    P = I^2*R = (E^2)/R = E*I

    p = (R*A)/(L)

    P[watt] power; I[amper] current; E[volt] voltage\electromotive force; ρ[#NA] resistivity; R[ohm] resistance; A[cm] cross-sectional area; L[cm] length;

    Maybe you were thinking about https://en.wikipedia.org/wiki/Joule_heating#Reason_for_high-voltage_transmission_of_electricity [wikipedia.org] ? It doesn't really matter here...

    --
    compiling...
    • (Score: 2) by Scruffy Beard 2 on Tuesday August 09 2016, @10:06PM

      by Scruffy Beard 2 (6030) on Tuesday August 09 2016, @10:06PM (#385991)

      In a modern computer case, 12V *is* the high-voltage line designed to reduce transmission losses.

      The CPU draws close to 100 Amps at about 1V.

      If you can make the "high voltage" line 48V instead of 12V, you can reduce the current.
      I=P/E

      100W @48V: 2.08 Amps
      100W @12V: 8.33 Amps

      Ratio: 1:4.01 (possible rounding error there)

      • (Score: 2) by RamiK on Friday August 12 2016, @01:39AM

        by RamiK (1813) on Friday August 12 2016, @01:39AM (#386858)

        I didn't mean you miscalculated ohm's law. I assumed some familiarity with datacenters... Well, lets break it down: What they want is to downscale the 2U 48v aux supplies \ combos into the 1U rackmounts.

        That's to say, they want to use: http://www.nipron.com/product_detail/index.cgi?p=20318401 [nipron.com]

        The "just wrong" part referred to the applicability of 48v for personal computers. That's to say, if we're following the formula with the real hardware figures, we can even see a x16.5 times size reduction in favor of 48v:

        P = I^2*R = (E^2)/R = E*I

        p = (R*A)/(L)

        p(L/A) = R
        P = I^2*R -> R = P/I^2
        ---
        p(L/A) = P/I^2 = I*E/I^2 = E/I
        L/A = E/(p*I)

        A = L*p*I/E

        The length and resistivity being constants between the two, they're omitted. -> A = I/E

        12v/16.5a: A = I/E = 16.5/12 = 1.375

        48v/4a: A = I/E = 4/48 = 0.08333

        That is, the 48v line is x16.5 smaller.

        -> However, a single 200w line is "just wrong" to 90% of users. All it will do is move PMIC into the motherboard where heat is harder to displace.

        And that's where my original comment started: It's a real concern that, following ECC RAM or datacenter class GPUs, we'll start seeing either more segmentation, increased prices or both with the introduction of 48v 1Us.

        --
        compiling...
        • (Score: 2) by Scruffy Beard 2 on Friday August 12 2016, @03:43AM

          by Scruffy Beard 2 (6030) on Friday August 12 2016, @03:43AM (#386889)

          I think I want one of those supplies.

          Especially useful if I ever get a 48V solar installation going.

          • (Score: 2) by RamiK on Friday August 12 2016, @01:38PM

            by RamiK (1813) on Friday August 12 2016, @01:38PM (#387007)

            I think the biggest ATXV48* you could get will have 0.7kw 48v rail and those panels are in the 2 - 7kw ranges... When it comes to solar, I personally like the idea the air conditioning companies are running with: They offer installs with dedicated solar support. So, you can just setup the AC to turn on whenever it's getting power off the panels and target a specific temperature, and then turn it on manually when it's cold and the skies are cloudy. That covers 2/3 of the year for me. Maybe even much of the winter :)

            --
            compiling...