How'd your servers get that baby-smooth look? Dutch and Brit cool kids dunk Supermicro systems in synthetic oil Dutch liquid cooling specialist Asperitas and British systems integrator Boston have linked arms on a range of server systems based on "immersion cooling" – with the hardware submerged in giant tubs of synthetic oil, a dielectric compound very similar to Vaseline or baby oil. Four fully contained standalone systems were …
I don't think it's significantly more magical than lift it out and try not to drip oil all over the floor.
I've seen demos where they'd more or less taken the top off a 1U/2U server and removed the fans then dunked it in a vat of oil.
Apparently ssds/helium hdds are happy enough being submerged too. Although I imagine it'd invalidate any warranty
At most datacenters I've worked with entire boxes are swapped out - the broken one taken away for further examination/repair/exchange/disposal. As long as they have nice self-sealing connectors for the plumbing they'll be fine.
The boxes will be substantially more dense than normal ones though - the oil will be heavier and they appear to be packing a lot more components into the same volume than a vanilla box... Perhaps professional wrestlers could do the job - they should be well practiced in grappling with heavy oily units. :)
> thanks to their self-contained nature, such systems can be deployed in any kind of building, in any climate – no data centre or dedicated server room required.
The heat still has to go somewhere. If the oil bath doesn't go through a heat exchanger then the heat is still going to warm the room, just a bit more quietly than if it had fans.
> ...and a secondary warm water circuit ejects the heat from the system" Isn't this a heat exchange?
You're right of course, but that doesn't really fit with their 'self-contained' and 'any climate' statements. If they'd just said "needs less intrusive cooling", or something of that ilk I'd be fine with it.
TBH, I can't see my clients with traditional DCs buying this - if you've already got all the air con why would you then even consider adding more complexity to your cooling arrangements? And if you haven't got a traditional DC why wouldn't you use the cloud?
If you had a lot of these, I would think the best solution would be a direct exchange system, where your HVAC has a "freon" (whatever the current substitute is called) loop piped to each cabinet.
The bigger problem would appear to be weight, though if you designed your datacenter around this you could dispense with the raised floor and it wouldn't matter if you had 3-4x the typical per sq ft weight.
This post has been deleted by its author
But not the hideously expensive Fluorinert used as the medium.
I do sometimes wonder if Cray thought about using purified mineral oil at around 1/1000 of the cost and then thought, no, expensive computer needs expensive cooling system, and if everyone else thinks fluorinert is essential, they won't copy us.
I was going to say that the Fluorinert couldn't be huge next to the total system cost, but then I looked it up and a Cray-2 was "only" $15M. Seems cheap next to today's massive clusters!
Still, a 750lb barrel of Fluorinet is "just" $35k. Old Seymour didn't believe in doing anything if it wasn't all-out. Bless him. His biography is fascinating.
https://www.chemworld.com/Novec-FC-770-Fluorinert-Electronic-Liquid-p/novec-fc-770.htm
Back in the early 1980s I worked on a R/F amplifier that was cooled by purified water. The water had to be ion free because the amplifier tube that was being cooled was sitting at around 7.5Kv. It worked well enough except in frosty weather -- the water froze when the prototype wasn't running.
(I've used a similar setup on load resistors I used for testing high power audio amplifiers. Nothing sophisticated here, I just dumped them in a bucket of water.)
I'll supply the mineral oil (baby oil), canola, corn, sesame, peanut should work also. Besides, think of the aromas!
Now to come up with a workload to put my demo units through their paces. Perhaps "Answer to the Ultimate Question of Life, The Universe, and Everything". Shirley with these specs the "Anna AIC (CLOUDzone243) is intended for GPU-based workloads like 3D rendering and machine learning – featuring the same 1,056 Xeon cores, but adding a whopping 331,776 CUDA cores across 72 integrated Nvidia cards" should be able to compute 42 in less than the 7.5 million years (probably performed on 8-bit CPUs.
Anybody have a spare hydro plant to loan me for a few million years?
Not sure if this is comparable to 3M Novec as this sounds like a traditional liquid cooled system compared to Novecs novel boil off heat transfer method. I imagine the price differential would be similar to the performance differential between the two solutions. Either way its cool to see this tech making its way to our datacenters.
Other times this has been tried it usually didn't work too well - poor circulation/localised cooling, reactions between components and the oil and issues with dielectric effects all tended to cause problems.
Also someone mentioned hard drives - they don't work for long either as either the pressure equalisation valve or a seal will give out and let oil in.
It's not that immersion cooling doesn't work at all, but you can't do it cheaply and expect a lot of success.
IIRC I read that in WW1 castor oil lubricant (used in aircraft engines so crude I just cannot image someone taking off in a plane powered by them) was inevitably taken in by the pilots in the exposed cockpits of the day, causing the runs.
The cure for the runs was whisky.
Part of the death rate was due to drunk pilots.
The thought of a data centre full of IT staff either headed for the loo or on the whisky bottle is a little concerning.
Hi all, we noticed a lot of questions around this article.
It’s impossible to answer all individually, that’s why we’ve planned a webinar on Friday, April 12th, 2019.
During this webinar Asperitas and Boston will cover all your questions. Register for free at https://eveeno.com/the-future-is-now
A long time ago, Shortly after the war, in fact, I read an article where the principle was tried in America. It was better than the system proposed here, in that it used a low boiling point fluid, One of the Freons, I believe. The advantage here is that taken away from the housing, the coollant would merely evaporate away, leaving the kit clean and dry. There was too the advantage of a boiling coolant, that the temperature of any part of the assembly could never achieve a temperature in excess of the boiling point, and the coolant didn't have to be pumped around the cooling circuit.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
