A humble proposal to Bitcoin miners

Thursday, 06 November, Year 6 d.Tr. | Author: Mircea Popescu

First, an illustration (snipped off Qntra and Kako) :


In this grand epic of ours, the interval between Beowulf above and Aethelbald below stretches just outside of a season, maybe five months. Apparently a hundred petahashes later, heat damage and fire-catching mining rigs are still the main problems of Bitcoin mining. Funny how that works, isn't it ?

To perhaps remedy this issue, definitively, here is my humble proposal : fill the fucking warehouses with technological gas, yo!

Specifically : seal up your warehouse, pump it full of Pentafluoroethanei and there you go. Place some compressor units outside to extract the heat, some fans inside to distribute the freezing winds and forget about the entire story.

With this arrangement you can in point of fact run your farm as far below zero as you wish - since the atmosphere is dry you'll never have condensation or ice formation. Simply insulate the outside in as many layers of polyurethane foam as you can bother with, three meters thick or whatever, and freeze that mofo. Not only do electronics work slightly better below zero, but should a compressor unit fail you have enough stored cool in there so the rigs can mine full blast for a few hours while you replace the thing. Never ever ever have any downtime ever again due to heat considerations, no matter what happens.

With this arrangement you can also forget about catching on fire : it simply won't ever happen, as there's no oxygen in there. Even should your datacenter somehow surge to 500 degrees : cut power, wait for a few days for it to cool, whatever melted melted but nothing burned, for absolute certain and ironcladly guaranteed. Look at that picture up top again and tell me this isn't worth something.

Sure, you won't be able to have people derp up and down the corridors, unless you put a lock in and buy them astronaut suits.ii Guess what ? They shouldn't be derping around your miners anyway! And besides, astronaut suits are pretty cool, a lock is not really that expensiveiii and if not you can simply turn off the farm, pump the technological gas into tanks, pump atmospheric gas in there, send the crew in for maintenance, pump air back out, HFC-125 back in and you're ready to roll. With some management the whole thing should be less than six hours. Look at that picture up top again and tell me six hours is a big deal.

And what about costs ? Suppose your warehouse is one hectare (~100k sq ft) in a 4:1 arrangement (200 x 50 meters) with ceilings at 7 m (23 feet, more or less standard). That means your warehouse contains 70k cubic meters of gas maximally. Adding a 5% technological margin (for whatever is in pipes, being compressed, whatever) and a 10% safety margin (so you keep some on the premises just in case you develop a loss, some in transit being shipped in etc etc) the grand total is 80k cubic meters of HFC-125 you could possibly need.

HFC-125 goes for about $3`800 per ton FOB. Since it's a gas, each kmol counts for 22,4 cubic meters, and since its formula is CF3CHF2 each kmol weighs 19 * 5 + 12 * 2 + 1 = 120 kgs. Therefore, the cost of your 80k cubic meters should be no more than 4`000 * 120 / 1`000 * 80`000 / 22.4, or about $1`700`000. Which yes, is more than a cup of coffee, but otherwise comes to about two dollars per sq ft, probably less than your monthly rent!

As a gas HFC-125 has a caloric capacity of about 0.85iv (water is ~4, air is ~1), which means that it heats about 15% faster than air. Your 70k cubic meters will require 850 * 70 * 70000 / 22.4 = 185 MJ to go from -10 to +60 degrees, which means 18kWh per square meter going for a full hour - roughly double what your power density is going to be. Even accounting for gradients is not going to change much (seeing how hot gas will go towards the ceiling, whereas your equipment is on the floor) ; meanwhile you can extract heat at any speed you choose, with that much gas to work with you're really only limited by how much you're willing to spend on compressor and heat exchanging elements. 1 TJ/h or whatever you wish, it's yours.

Polyurethane foam costs about $200 per cubic meter. If you go crazy and insulate 1 meter thick you'll need (50 + 200) * 2 * 7 + 3 * 10000 = 13`500 cubic meters, coming to about $2.5 million. It has a λ10v of about 0.02 (more or less the same as air), which means that one meter's thickness will ensure a total power drain out of your construction of just about 300 Watts, very roughly speaking. Not the worst of your problems.

So to sum it up : for ~2mn for the gas and ~3mn for the insulation, plus maybe 1-2mn more for various trimsvi you'd have 10k sqm of heat-inert floor. < 1k$ per square meter should not be a consideration, seeing how this arrangement will likely exceed the lifetime of your miners, and seeing how you can now pile up at least five meters' height of mining gear, and you can put 1 MWh through each of those square meters of floor without a care in the world. Anything you buy should be doing at least 1Gh/J, which means each sqm should do at least 1Ph, which means that a) your datacenter so equipped should be able to support 10 Yh, roughly 100x total network output currently, and b) your ~$700 or so cost per sqm of flooring amortises in all of... wait for it... o my gawd... 25 BTC per block, $250 per BTC, 1% of total output, one block every 10 minutes... can it be... two hours ?!vii Oh dear!

And guess what : if you're not really planning to deploy 10 Yh, you could just as well scale the foregoing down a bit. A 10 x 10 meter room for a notional 1/100 floor surface would, with ceilings at 2.75, only displace 275 cubic meters, a mere 0.4% of our original 70k. Which means that yes, the ~$700 per sqm will prolly go up, though not likely to 1k, whereas the total cost would go down, though not likely all the way to... $40k ? You mean less than a fucking car, let alone jets and whatnot ?

Seriously people. Stop thinking with your girlfriend. There's absolutely no need to hear about how some schmuck bought himself a twenty foot gilded dildo encrusted with Bohemian lead crystalsviii. There's high need of hearing about some sane miner made the first-ever HCF-125 submerged datacenter.

Get on with it.

  1. It's an excellent choice for this application because it's a refrigerant gas so inflammable it actually doubles as a fire supression system. Seriously, if you sprung for the best fire supression money could buy you'd basically have a system that dumps CF3CHF2 installed anyway. And also, having exactly zero ozone depletion factor and being the replacement of choice all over the US for older refrigerants, you can advertise your environment-friendliness with excellent basis in fact (it won't matter for squat, because the envirofreaks are not even slightly fact based, they just hate capital and industry - but that's an entirely different discussion). []
  2. Mostly so they can continue breathing - otherwise HCF-125 is not toxic nor irritating - it's pretty much copacetic at best. []
  3. What, two interlocking steel doors and a pump ? O noes! How could this be ever made!!1 []
  4. Here's a full discussion of the matter. []
  5. W / K / m []
  6. Industrial compressors are really cheap when compared to all this, piping's also not that much of a big deal. []
  7. 700 / (25 * 250 ) * 100 * 10 / 60. []
  8. What the fuck do you think "swarowsky" is ? Back when I was a child the house was fucking full of Bohemian glass, my mother still has a closetfull of that crap, vases and fruit bowls three to five kilograms a piece. If I start polishing the pile into half carat "crystals" are you going to buy it from me for twenty billion dollars total ? Idiots. []
Category: Bitcoin
Comments feed : RSS 2.0. Leave your own comment below, or send a trackback.

15 Responses

  1. Aside from the intricacies of engineering a pressure vessel large enough to hold the goods (approximately reduces to building a... submarine. why not use existing submarine?) let's have the real boojum:

    Thermal conductivity of the gas. Mechanical refrigeration works on account of 'PV == nRT' - but where does the cooling actually take place? Where gas expands. That is, where the 'hot' line from compressor station enters the chamber. You can work out the flow (good exercise) or simply visit a facility with large walk-in refrigerated warehouse and ask how it is put together and why. (Yes, this superficially sounds like 'argumento ad derpulum' but mechanical refrigeration is fairly dusty and mature technology.)

    So where's the real ticket? Miners welded into sealed 'hedgehog' capsules submerged in water. If 'caviar' budget, 'inconel' hedgehogs, deep salt water. If 'pizza' budget, steel hedgehogs, submerge in freshwater pond that is to be heated anyway (e.g., fish farm.) No fire concerns, one flooded capsule is merely one lost processor and its fuse blows. And the piranhas will molest unwelcome underwater guests.

  2. Oh, and power the hedgehogs inductively through chassis. Comms through same or through ultrasonics.

  3. Mircea Popescu`s avatar
    Mircea Popescu 
    Thursday, 6 November 2014

    This is not pressurized. 1 bar.

    Nothing prevents you from piping the liquid gas all the way into the warehouse, then release it out of floor placed valves.

    The fish farm / submerged miner idea however is also pretty good.

  4. Incidentally, a more interesting (and actually practical) version of your apparatus would simply spray pressurized CO2 (perhaps mined from atmosphere in the first place using, e.g, zeolite) onto the heat exchangers. (You still need Al or Cu heat exchangers, because chip packages have finite temp gradient ratings, esp. under varying thermal output.) The CO2 can simply be allowed to fill the room. Unmanned building, naturally.

    The underwater hedgehogs have the arguable merit of actually being cheaper (total cost of ownership) than present-day state of the art (there are ponds that need heating, but not quite sufficiently many fissile power stations!) Not only avoids fire, resists thieves, but simplifies shelving! Just clamp on the inductive cable and drop the thing in the tank. Let it finds its own place. No dust to sweep, likewise. If a particular hog gets 'crowded' and begins to 'feel too warm' (internal sensor) just yank on the tether a bit.

  5. Mircea Popescu`s avatar
    Mircea Popescu 
    Thursday, 6 November 2014

    Meh practical my foot. Do the math, see what the extractors would run you, and then how much they'd cost to run.

    Anyway, the heating fisheries thing is probably inferior to the ancient b-a patent of "ceramic tile/elements with embedded miners". Which will definitely happen eventually, but is an entirely different industrial consideration. What's in discussion here is maximal concentration, not ubiquity solutions.

  6. Peter Lambert`s avatar
    Peter Lambert 
    Thursday, 6 November 2014

    This is an interesting idea.

    Perhaps start with a server rack sized container which is sealed and filled with the coolant gas?

    How about using a shipping container?

    To build on Stan's idea, you could even dump the shipping container into a nice cold lake, perhaps Lake Superior?

  7. Mircea Popescu`s avatar
    Mircea Popescu 
    Thursday, 6 November 2014

    Shipping container is not really built to go into water.

    There already exists a company that markets exactly this, Ro-Ro containers filled with miners immersed in coolant. I think they use liquid phase tho (it'd make sense, solid steel is basically one large canister, much better structural integrity than a proportionally much thinner concrete wafer ; liquid thermic characteristics both for capacity and exchange much better than any gas).

  8. MP:

    > ceramic tile/elements

    Works, but tiles are a bitch to swap out. Cement, grout? Screw it.
    Hedgehogs, let'em sink. Or float, for that matter. Like naval mines.


    > filled with the coolant gas

    Gasses have terrible thermal conductivity! They obey 'ideal gas law' and can be used to move heat (refrigerator scheme) - yes. Gotta recall the kindergarten basics, we're talking 1830s state of the art here.

  9. I suspect shelf lifetime of mining hardware is so short, that it needs constant replenishment with newer units. This may make sealing impractical.

    Maaaybe when we're finally on 10nm process..but even then it's possible that letting 5 chips burn is cheaper than properly sealing/cooling one.

  10. Mircea Popescu`s avatar
    Mircea Popescu 
    Friday, 7 November 2014

    @Stanislav Datskovskiy Nobody ever takes them out, they're heating elements.

    @hurov This "letting 5 chips burn" sounds a lot like "letting 5 workers strike". If only it worked that way.

    As discussed, downtime due to sealing should be ~6 hours more than downtime otherwise. This is not significant, inasmuch as you'll rarely upgrade a DC without 6 hour's notice.

    I scarcely can see what 10nm process can possibly have to do with it, most Z80 processors made in the 70s still work.

  11. GAW is a scam`s avatar
    GAW is a scam 
    Friday, 7 November 2014

    75% of the 13500 cubic meters of insulating material are used for the ceiling of the datacenter. You could save a whole lot of money by simply installing a pool instead. If it rains the water is free, but even if it does not rain it doesn't cost 200 dollars a ton. Plus you can make it 3-5 meters deep, there is a lot of cooling capacity there.

  12. The average nuclear power plant produces something like 1GWh per unit these days. At 1MWh per m^2 of floor space total power needs to run such a DC would be 10GWh. Enough to suck dry 10 units. There is no way anyone anywhere provides you with that much power, so this idea definitely depends on getting the scratch together and the approvals needed to build a huge nuclear power plant on-site. It would be huge : all of Belgium produces under 6GWh, Canada produces under 10GWh, this would be something of the size of China's Haiyang I, II and III combined which they haven't even built yet.

    However, there would be very good synergies with nuclear plants on site. You could dig the datacenter into the ground rather than having an above ground structure to house it. This would muchly improve your insulation. Placing the nuclear power plants atop 500 tons of HCF-125 kept cool would be a great idea, both for the emergency cooling available through purely mechanical means that can very well be a life saver, and also for the fire supressant properties which idem. All of a sudden it makes sense to buy not 500 tons but ten thousand tons and just keep it there in case there's a malfunction of any kind. If Fukushima had your datacenter under it the Fukushima disaster would have been avoided, at the cost of some lost mining power over a few days.

    Another big deal is the heat exchanging. You could use the 1TWh waste heat coming out of the hot side of your compressors to pre-heat the water used in the power plant turbines. This alone would boost efficiency to levels not heard of. Coupled with the fact that because of the geographic arrangement you don't need transformers for transport upping of the current produced, have no wire network to support etc, and the fact that nuclear power is the cheapest power outside of hydro (because of economies of scale the 10GWh plant will probably make electricity cheaper than any hydro plant), this plan would totally kill everyone.

    It would also cost more than a small country, what with the billions of dollars worth of miners and the billions of dollars worth of nuclear gear. Nevertheless...

    Good plan!

  13. Mircea Popescu`s avatar
    Mircea Popescu 
    Friday, 7 November 2014

    @GAW is a scam Yes it's a scam lol.

    Anyway : maybe the water is a good idea, but it has some significant problems.

    One is that while blessed with a lot of caloric capacity, water also has a significant conductance factor. So, the polyurethane provides a lot of thermic insulation, a large pool of water provides a lot of thermic inertia. These are not at all the same, to better understand this consider the problem of road skidding. If the road is icy, throwing sand on the ice provides a benefit similar to better insulation in our case, whereas increasing the size of the vehicle provides a benefit similar to what you're proposing. Intuitively it's obvious larger vehicles aren't inherently safer, which well illustrates the problem.

    The other is that water and electronics don't well mix, and putting a whole lot of water on top of a whole lot of expensive electronics is mechanically unsound, any failure (doesn't even have to be an earthquake) will hit you right where it hurts. This is exactly the sort of bad design that Taleb tends to rail about, and it happens to also be exactly the sort of bad design that caused - or contributed to - Fukushima.

    That said, there may be merits to your idea, especially if in a location with good precipitation, where the water cools significantly at night and if you're getting a significantly lower power rate during nigth time (as is often the case around large towns). You could run your thing full blast during night time, throttle it in the day and realise overall better efficiency, perhaps. Kind-of iffy though.

    @Anon I am in awe of the realisation of what you're saying there. Yeah, totally, the pre-heated steam is happening.

    Sadly, I am also in awe of the realisation of the actual scale of this project. You're right, as designed a hectare of datacenter really needs many square miles of power plant, reducing the entire problem to a very different arrangement willy-nilly. In my defense, I was using using the hectare example to make a point.

  14. > conductance...

    I'll add that it is entirely practical to build a miner (as in chip package) that laughs at water. Power and comms (no serious bandwidth needed, if you recall) through inductive loop.

    This is distinct from the earlier 'hedgehog' proposed, in that there is ultimately no naked conductor in the mix at all. (No pcb in the usual sense, etc.) Just a segment of glass pipe with insulated cable wrapped around it.

  15. Mircea Popescu`s avatar
    Mircea Popescu 
    Friday, 7 November 2014

    Now the glass pipe arrangement may really work. Build datacenter as a tube, run water through the tube (or put on river bed).

Add your cents! »
    If this is your first comment, it will wait to be approved. This usually takes a few hours. Subsequent comments are not delayed.