From: cr88192@gmail.com   
      
   On 9/25/2025 9:03 PM, Scott Lurndal wrote:   
   > MitchAlsup  writes:   
   >>   
   >> Terje Mathisen  posted:   
   >>   
   >   
   >>> I am quite sure that number is simply bogus: The power factors we were   
   >>> quoted when building the largest new datacenter in Norway 10+ years ago,   
   >>> was more like 6-10% of total power for cooling afair.   
   >>>   
   >>> . a quick google...   
   >>>   
   >>> https://engineering.fb.com/2011/04/14/core-infra/designing-a   
   very-efficient-data-center/   
   >>>   
   >>> This one claims a 1.07 Power Usage Effectiveness.   
   >>   
   >> All of this depends on where the "cold sink" is !! and how cold it is.   
   >>   
   >> Pumping 6ºC sea water through water to air heat exchangers is a lot   
   >> more power efficient than using FREON and dumping the heat into 37ºC   
   >> air.   
   >>   
   >> I still suspect that rectifying and delivering clean (low noise) D/C   
   >> to the chassis' takes a lot more energy that taking the resulting heat   
   >> away.   
   >   
   > The FB article above describes how they reduced the   
   > losses due to voltage changes as well as rectification.   
   >   
   > Consider that there are losses converting from the   
   > primary (e.g. 22kv) to 480v (2%), and additional losses   
   > converting to 208v (3%) to the UPS.  That's before any   
   > rectification losses (6% to 12%).  With various optimizations,   
   > they reduced total losses to 7.5%, including rectification   
   > and transformation from the primary voltage.   
   >   
      
   Hmm...   
      
   Brings up a thought: 960VDC is a semi-common voltage in industrial   
   applications IIRC.   
      
      
   What if, opposed to each computer using its own power-supply (from 120   
   or 240 VAC), it uses a buck converter, say, 960VDC -> 12VDC.   
      
   Or, 2-stage, say:   
      960V -> 192V (with 960V to each rack).   
      192V ->  12V (with 192V to each server).   
      
   Where the second stage drop could use slightly cheaper transistors, but   
   still limiting electrical losses due to wire resistance, while still   
   avoiding losses due to transformers and rectifiers.   
      
   To balance cost and efficiency, could use, say, 8 or 10AWG CCA (copper   
   clad aluminum) vs 10 or 12AWG copper. Could run the wires at a   
   relatively lower amperage rating, say:   
      8A over 10AWG CCA   
      16A over 8AWG CCA   
   Or, roughly 1/3 nominal.   
      
      
   Where, CCA wire is a lot cheaper than copper wire, so it is easier to   
   justify using absurdly thick wire here.   
      
   Where, contrast say to running 8A over 20AWG, which works, but a fair   
   bit more is lost due to heat. Or, the alternative could be to run the   
   power over parallel thinner wires rather than a single thicker wire. For   
   example, replacing each 10AWG wire with four 14AWG wires.   
      
   8A at 192V being 1.5kW, and 8A at 960V being 7.7kW.   
      
      
   Though, assuming a series of 16 racks running on each shared 960V bus,   
   this would be 128A. The above de-rating scheme would likely make normal   
   CCA wire impractical. Probably could distribute DC power over a pair of   
   1.25" aluminum bars or 0.75" to 1.0" copper bars. Likely, the 1.25"   
   aluminum bar being the cheaper option here.   
      
   Could maybe then connect each 10AWG wire to the bars using a clamp,   
   and/or use an intermediate socket or modular connector.   
      
   Does kinda seem a bit overkill though.   
      
      
   Main power distribution would likely need to operate at a higher   
   voltage, otherwise the building-scale power rails would be absurd here.   
      
   Say, if one assumes a monolithic 960VDC system, and 16 rows, this is   
   2048A. Like, what does one do here, 3" copper or 5" aluminum rails?...   
   Probably no.   
      
   Well, or maybe get creative and use large aluminum I-beams that serve   
   both as power distribution and joists (so, all this metal can serve   
   additional purpose). Though, 960V through the joists seems like a   
   building maintain maintenance hazard. Say, for example, 0V through the   
   floor and 960V through the ceiling.   
      
      
   Input power would likely need multiple transformers and rectifiers to be   
   practical; though admittedly I have little idea here what sorts of   
   diodes would be used in these rectifiers. Seems like each diode would   
   itself need to be stupidly large to deal with this crap.   
      
      
   As for cooling, could maybe either use liquid cooling, or hybrid   
   aid/liquid (say, with superchilled liquid pumped through radiators, and   
   then fans circle air through these radiators).   
      
   To move lots of heat, could maybe use -90C ethanol as a coolant. Where   
   ethanol can be pumped like water, but could be nearly as cold as Freon.   
   Would likely still need big refrigeration pumps.   
      
   If one could have an artificial lake outside (preferably with a   
   sun-blocking cover), this could be used as a heat-sink.   
      
   Where, say:   
      Inner loop uses cold ethanol;   
      Refrigeration system moves heat from ethanol loop to a water loop;   
      The water loop pumps to/from an artificial lake used as a heat sink.   
   If the lake is above ambient, it will dissipate heat, but if too much   
   higher it would suffer evaporation looses.   
      
   One idea here could be to have 2 levels of cover over the lake:   
   The lower one is a metal cover painted black on both sides, placed   
   roughly 20 inches over the surface of the water;   
   The second cover is another 20 inches higher, painted black on the lower   
   side and white on the upper side;   
   The lower cover has a blocking wall to limit how much water vapor   
   escapes, whereas the upper barrier is open to the sides (allowing air to   
   flow through).   
      
   As the water evaporates, it moves heat into the barrier, which then   
   radiates heat (as black-body radiation) where the water condenses and   
   falls back into the lake;   
   The upper barrier partly absorbs heat from the lower layer, and also   
   serves to reflect the sun. Air-flow between the layers can be used to   
   radiate heat.   
      
   One other possibility being to have a tall tapered tube (narrower near   
   the top) with an open top, with the coolant water in the bottom (with   
   the tube tube serving to reduce evaporation loss, as water is more   
   likely to re-condense on the walls and fall back down than to escape the   
   top). Could likely be made out of steel or similar, maybe black inside,   
   white outside. Then maybe could heat the coolant water to around 70 or 80C.   
      
   While in theory, a giant radiator could work, a sufficiently large   
   radiator would likely be impractically expensive.   
      
      
   Well, don't know what people actually do, this is just what comes to   
   mind at the moment.   
      
   ...   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)