Super-cooling liquid cuts 97% off data centre cooling costs | Hacker News Reader

Super-cooling liquid cuts 97% off data centre cooling costs

theregister.co.uk

99 pointsiProject47 comments

It would be nice if there was a page that covered this without (a) calling us "boffins" and (b) missing the major point that the coolant boils and (c) being full of factual errors, as you'd expect from an article that uses the term "boffin". (After all, if you're smart enough to get the basic facts right on a story like this, you're probably a "boffin" yourself.)

But http://blog.iceotope.com/2012/03/iceotope-liquid-cools-cabin... omits the crucial detail that the Novec boils inside your server, which is why it can work so well, and http://solutions.3m.com/wps/portal/3M/en_US/NA-DataCenters/D... of course doesn't mention Iceotope at all.

(Thanks to uvdiv for digging up the actual information!)

"boffin" is British slang for a scientist/researcher. It's got generally positive connotations indicating that the said person is smart.

Since The Register is a British site, and is also known for often being somewhat tongue in cheek, there really isn't anything to read in to their use of the word.

http://en.wikipedia.org/wiki/Boffin

Still, science reporting is generally terrible and the Register is an awful awful source.

From the refs on that page, http://www.guardian.co.uk/science/blog/2010/sep/24/scientist...

"There is little that irritates scientists more than the idea of the "boffin". This century-old meme has at least two flavours: the befuddled, bespectacled, bad-hair-day (or no-hair-day) man, socially inept but somewhat cuddly (think Doc in Back To The Future); and there is the more sinister iteration: the equally dishevelled but cold, arrogant and/or mad male meddler, bent on no good (think Rotwang in Metropolis).

"Neither of these versions is remotely flattering, and neither bears any resemblance to reality."

More details:

(press release) http://solutions.3m.com/wps/portal/3M/en_US/NA-DataCenters/D...

(technical data) http://solutions.3m.com/wps/portal/3M/en_US/3MNovec/Home/Pro...

It's a phase-change system (the coolant boils). That's what the article means by "a thousand times better at conducting heat than water": it absorbs orders of magnitude more heat because of the phase change. It boils 49 °C at atmospheric pressure, cold enough for electronics.

They claim it doesn't need fans: the boiling is enough to maintain natural circulation (convection) from the server to the heat sink.

Can you explain how this is different from fluorinert?

Hopefully cost, fluorinert is vary expensive. Also, with a low boiling point of 49c you should be able to avoid pumps as long as you don't mind the components sitting at 120f.

This should be comparably costly, it's also a perfluoro compound. C2F5-(C=O)-CF(CF3)2

Wouldn't it be pretty toxic (especially if it's a vapour?) In a sealed component in a datacentre that problem doesn't matter, but they're speculating about home use also.

I don't know about that compound being toxic, but there are some splendidly awful fluorine compounds:

http://pipeline.corante.com/archives/2010/02/23/things_i_won...

Though that's completely different, it's an oxidizer.

You have fluorine in your tooth paste. Hell, oxygen can be dangerous, nitrogen is used in a lot of explosives etc etc.

Probably not, C-F bonds are extremely stable, so perfluoro compounds are biologically inert. The datasheet says they made some animal inhale 10% (!) Novec gas for 4 hours without effect.

http://multimedia.3m.com/mws/mediawebserver?mwsId=66666UF6EV...

The idea of liquid breathing diving systems is usually tied to a perfluorocarbon.

It has half-life of 5 days in the atmosphere, so it's not that stable. According to the MSDS it and degrades both when exposed to UV and at high temperature (like those present in a structural fire) into hydrogen fluoride gas, which becomes hydrofluoric acid upon contact with moisture (including biological tissue).

Also you're supposed to dike the area if a major spill occurs, but I'm pretty sure that's just MSDS boilerplate.

No, fluorocarbons in general are very inert, with the astonishing result that they're generally less toxic than water.

The 3M leaflet suggests that you should avoid breathing it, or getting it on your skin or in your eyes, but that you shouldn't be harmed if you ingest it.

I don't know if that's relative to some of the other nastier stuff they sell.

The 3M page seems to show servers dunked in "semi sealed" cabinets of this stuff, which boils, evaporates, cools, and falls back into the tank.

I am not an expert in the field, but a quick look through the product pages at 3M suggests that the important difference is the boiling point. Novec boils at 49 degrees C, so for cooling electronics the primary mode of heat transfer will be the phase change. Fluorinert seems to boil at higher temperatures, 80 and > 200 degrees for the two variants I looked at.

They seem to be in the same family of products and several of their physical properties are very similar, so maybe they're just variants.

Here's a link to a page which has both families:

http://www.mgchemicals.com/products/cleaners/3m-fluids/

I have no knowledge on this subject, but

- According to wikipedia [0], fluorinert as used in the Cray was a single-phase coolant (no boiling). It needed fluid pumps and chilled water, at high energy cost; this phase-change solution avoids both.

- Wikipedia claims fluorinert has a very long atmospheric lifetime and high global-warming potential. [1] 3M claims this new fluid is destroyed by sunlight [2], and doesn't have the same problems.

[0] https://en.wikipedia.org/wiki/Cray-2#Packed_circuit_boards_a...

[1] https://en.wikipedia.org/wiki/Fluorinert#Toxicity

[2] (PDF) http://multimedia.3m.com/mws/mediawebserver?mwsId=66666UF6EV...

(The "M.J. Molina" in the datasheet is the Nobelist and co-discoverer of the ozone hole)

I'm confused, or maybe The Register's article is. It states "boffins at Leeds University and British start-up Icetope have invented a super cooling liquid", but also refers to "Dunking servers in new magic liquid 3M Novec" and has no mention of phase change/evaporation.

Are we talking about a single liquid? Did 3M license the formula from Leeds Uni/Icetope?

I'll try to understand it better this evening

I think I was wrong. There's only one coolant (or coolant family), but several server-cooling systems. There are both liquid and phase-change systems, and I'm not sure which this is.

3M is the chemical firm which developed the coolant ("Novec"). Leeds/Iceotope is a customer:

"UK company Iceotope, that patented liquid cooling for servers in the data center back in 2009, has released new products to market that use 3M’s Novec, an engineered non-flammable solvent it says makes the cooling process safer."

http://blog.iceotope.com/2012/03/iceotope-liquid-cools-cabin...

3M is also developing a server cooling system, independent of Leeds/Iceotope. This is the one in this press release, and I was mistaken to identify the one with the other.

http://solutions.3m.com/wps/portal/3M/en_US/NA-DataCenters/D...

The 3M system is phase-change. I'm no longer sure what the Register is writing about, but Iceotope is researching both liquid and phase-change systems:

"Iceotope Ltd, registered in Guernsey, is a technology company that researches and develops liquid and phase change cooling solutions based around heat pipe encapsulation."

http://blog.iceotope.com/2011/12/assets-of-green-data-centre...

Here's their liquid cooling system. One property of these coolants is their very low boiling point. That's useful for phase-change cooling. Another is that they have very high thermal expansivity, which means that they convect strongly as liquids (without phase change). That's useful for strictly liquid cooling.

http://blog.iceotope.com/2012/09/liquid-cooling-halves-data-...

They say the convection is 20-40x stronger than water, and overall heat removal 10-15x better. Since the Register article is saying "a thousand times" better heat conduction, I think they're taking about a phase change system. At least for that specific figure.

Thank you for digging deeper. Novec bears further watching. Iceotope are a bit odd, their website is just press release after press release.

Why not just use standard freons?

Freon is being phased out due to its ability to damage the ozone layer. I think you mean why don't they use standard commerical refrigerants. I believe the answer is that the 3M Novec family of engineered fluids has higher performance and lower toxicity/GWP.

Interesting coincident, just yesterday i watched a Russian tv show about dry water and they show that it doesn't conduct electricity and it has a low boiling temperature.

http://youtu.be/WIVt66RSWNU?t=9m31s (sorry, only in Russian)

Sorry for being a bit off topic, but "super-cooling liquid shaves" sounds like the intro to a Gillette commercial.

Phase change sounds a lot cooler then what the article was trying to put forth.

Dielectric liquid cooled computers have been around a long time. I even did my capstone project my senior year taking a 3M dielectric fluid and attempting to make a liquid cooled laptop.

What's always forgotten is you have to move the heat farther then you think. Dielectric fluids are great for getting the heat off the cpu but then where do you move it to? In my project we tried to use the back of the laptop monitor to be a large surface area for natural convention. Boiling is much cooler (literally).

Laptop of course wouldn't work out because imagine trying to manufacture those things at scale! Also, these fluids when they evaporate aren't exactly kind to the atmosphere.

You still need to dump the heat somewhere, no matter how efficiently you get it out of your server/cpu.

In colder climates heat is used for heating, you know? It's not 'dumped' anywhere. District heating networks are already in place in most countries, and it's trivial to connect the server farm in the heat network, possibly replacing a fossil fuel plant.

Heat isn't equal to heat. Server farms produce relatively low temperature heat, which is almost useless.

Look up entropy and thermodynamics for more background.

Odd word choice - 97% isn't really a shave.

Reminds me of when they first demoed CFC's on the program "Beyond 2000". Turns out that stuff ate the Ozone Layer. Makes you wonder about the unintended negative side-effects about new discoveries.

Makes me wonder about the unintended positive side effects of new discoveries.

Unfortunately, those are probably much less likely directly...

Awesome - this could be great for helping reduce cooling related energy use / slowing global warming, but control of a relatively benign chemical will be hard to force since and it's 280-320x worse than CO2 in terms of Global Warming effects if released....per MSDS http://multimedia.3m.com/mws/mediawebserver?mwsId=SSSSSuUn_z...

Sounds very cumbersome to have hardware adapted to use this, also expensive.

I'd still like to see something like this for server cooling: http://www.extremetech.com/computing/131656-the-fanless-heat...

Thanks. I missed that one. Not much cost info there but, I like the sound of:

"In computers, a Sandia Cooler would mean that we could finally cross the 4GHz/150W TDP thermal wall — or build computers that are thinner and quieter."

I think the idea behind showing an iPhone dunked into it is that the cost of adaptation is: buy vat.

Even if your only change was to place the server into the liquid you get a nice bump in efficiency of cooling by having the thermal bulk of the liquid transferring heat quickly away from the hot spots. The liquid itself would then cool on contact with the air, the much larger surface area acting kind of like a big radiator.

http://en.wikipedia.org/wiki/Heat_pipe

This is better because... there's no per-server hardware to get the heat to the cooler? Just a mostly-sealed room (or tank) with one inlet pipe and one outlet vent?

Uhm, that just moves the heat to another place... usually rather closeby. Not really a solution for datacenters (although they are used there, too, sometimes - but just to move the heat slightly further away from components so the main cooling can get to them). I suppose you could build them longer than you usually see them these days, but that would make them quite a lot more expensive.

Couldn't the cold end of one be attached to the hot end of another? I was thinking have a large one next to each rack, where the cold end goes to wherever the condenser for this system would be and the hot end connects to the cold ends of smaller pipes going to each server.

I'm assuming that tanks full of presumably patented special-purpose liquid would be rather expensive, and that sealed partially-evacuated piping ought to be somewhat less so.

Rather than assuming, why don't you check out their website (linked somewhere else in this topic)?

As for chaining up heat pipes... not really sure how to respond to that.

Because I assumed that this would be "call us to negotiate a price", which doesn't exactly help. And looking at the posted links that seems to be correct.

Heat pipes are passive, though - there's no pump to carry the heat as far as you need to.

HOpe this brings in new found joy to hosted SAAS/Cloud players who are competing with tight margins

Brings new meaning to the term wet-ware.

How is this different than dunking your machine in mineral oil?

It's a refrigerant. Boiling point is 40C at atmospheric pressure.

Reply on news.ycombinator.com