cooling

Power and cooling for free, the more power you use the more cooling you get, how neat is that? !

Not if you turn a can of compressed air upside down and spray the chips first, cooling them to -50C! He used this to recover encryption keys and defeat whole disk encryption.

* What's done for cooling/heating? Does this require a pretty temperate environment, or is it relatively well sealed?

I once read an article about a technique Intel had developed for improving cooling of processors by changing the shape of the fan. I related this to some of my co-workers.

Clearly there is going to be bandwidth at scale its like $2 - $3 per megabit per month, you've got power/cooling in a colo, so that's going to cost you maybe $3K/month for a 42U rack's worth of servers. And presumably you've got a couple of opsen types tending and feeding that rack.

However, JRuby still might be running slower than it would on a box with appropriate cooling. Right. Your computer can't be trusted to not throttle its performance due to crap cooling, and you're trying to benchmark using it?

Either way you're looking at 24 - 30 'circuits' for this space and those are probably about $500/month each so another $12-15K/month in 'power+cooling' charge. You pretty much have to add in either the cost of a tech or half the cost of one of your operations employees to run this setup.

I assume it's the chatty discussions which you're concerned about cooling down, so this would handle the problem case while avoiding the side effect on quiet/abandoned threads.

I think you can understand coffee cooling quite well without any quantum stuff - the atoms in the coffee are moving faster than those in the room. There will be a tendency when one impacts with an atom of the air in the room for that to speed up and the coffee atom to be slowed.

Diamond is a wonderful conductor of heat and so if you make your processor on a diamond subtrate you can pull lots of heat out into an attached cooling system. Get ready for 7" x 7" heat management assemblies though that attach to these things.

Fortunately, the engineering on these plants is done extremely carefully, with multiple independent systems in place to prevent exactly this by halting the nuclear chain reactions with neutron absorbing rods as well as by ensuring the flow of cooling water. However, the events in Japan exposed an "unknown bug" in the design, that if you flood the basements of the plants with water you can lose backup power and coolant flow.

When we switched to fluid-based cooling things changed dramatically. One of the design challenges was to maintain a narrow delta-T across the LED array. This is because thermal uniformity was required in order to have uniform performance across the array. The fluid solution, with some tricks, could easily achieve ten times better thermal uniformity than the air-cooled approach. And, in addition to this, cool the entire array to a much lower final temperature. A fluid cooling system was constructed using only a small fluid pump and no air-moving fans at all. A passive natural convection radiator could easily handle the heat-load in a normal air-conditioned office environment. While I have not looked at the specific case of cooling a CPU, based on my experience I have to say that far greater gains can be had by rapidly moving heat from the CPU surface using fluid-based cooling. This, effectively, creates the opportunity for much greater surface extension than can reasonably be applied to the small surface area of a CPU. Again, I have never studied CPU cooling, but I am not sure that this 150W cooling limit applies to fluid-based cooling. I can see building a systems that can very easily move 150W, or even double that, using a relatively simple fluidic cooler. At some level it is a matter of how many molecules of the fluid you can move across the CPU-side heat exchanger per unit time. The answer to that is "a lot". I can't see the Sandia or any other pure air-based cooling system used for CPU cooling at the extremes. The assembly would have to be very precisely manufactured and lots of work would have to be done in order to ensure that vibrations and harmonics of the motor drive system itself don't cause damage to the circuit board. If the system needs to have an impeller spinning at 2K RPM or more, lots of work needs to go into making it safe for servicing as a metal impeller like that can shred fingers in an instant. Finally, there's the question of the mass of the spinning impeller. In order to transfer heat into the impeller blades you are limited to certain geometry. If the spinning base and/or the blades get too thin you simply won't be able to move the heat out no matter how well it can move from the stationary plate up to the revolving disk. This is critical and it means that there are certain minimum geometry constraints that are likely to make the impeller somewhat massive. From my FEA work on heat transfer I know that you can only go so thin on blades before they become useless past a few millimeters above the heatsink base-plate. The same is the case here. What I can see is the use of this concept to create a fluid based solution that uses a liquid to quickly move heat from a CPU to a much larger heat exchanger that uses the Sandia heatsink to move heat into the surrounding air, and, thereby, cool the CPU. Even at that, I'd like to see data comparing conventional forced-air convection cooling of the external heat exchanger and even a comparison to a natural convection solution.