Castrol, Submer shift gears to datacenter immersion cooling Castrol, better known for its engine oil, has partnered with cooling specialist Submer to drive the adoption of immersion cooling for datacenter and edge applications. For those of a certain age, Castrol will forever be associated with TV ads that proclaimed its Castrol GTX product as not just oil, but "liquid engineering." …
If they are just using mineral oil, then I do not see any reason why people would not stick with good old fashioned Hydrofluoroethers (e.g. Novec 7000 AKA 99.5% 1-methoxyheptafluoropropane AKA C3F7OCH3 ). The linked video shows HFE's working well 10 years ago, should work just as well now.
The only thing I can see going for Castrol, is that their product (unless it actually is something totally new) should be much much be cheaper to distil from oil and eventually dispose of safely.
But on the downside, ever tried to get oil out of any electronics. In comparison a liquid that boils at 34°C (93.2°F) would be simple to remove.
Maybe Isobutene if using hydrocarbon as the source, for maximum fire risk (I'd hate to see the jump in insurance premiums) with a boiling point of −6.9 °C (19.6 °F; 266.2 K). Oh and lots of new and interesting problems with ice forming everywhere from the normal water vapour in the air.
Or 1-Pentene with a boiling point of 30 °C.
> ether's boiling point appears to be 35C, but unfortunately it's very flammable for that very reason
1-methoxyheptafluoropropane is Non-flammable, 100% ozone friendly, Non-corrosive, High electrical resistivity of about 10^11 Ω·m (roughly the same as glass).
Hydrofluoroethers are nearly all Non-flammable, their one big downside is the special handling required when eventually being disposed. This typically requires the product being shipped back to the manufactures and for them to handle all legal, environmental and safety requirements when dealing with poisonous by-produced like hydrogen fluoride and carbonyl fluoride produced during extreme heat thermal decomposition.
OK, so for a single system this might be nice.
But, scale this up in a busy datacentre with 100s of parts being replaced per week, there's going to be lots of parts dripping with fluid:
-As they are transported across the data hall after removal.
-In situ as they are being removed.
-In boxes or anti-static bags in preparation for logistics return.
-In storage areas as they await collection (pooling).
-In field engineers cars if the DC doesn't allow parts to be stored onsite (3rd party/co-lo DCs charge per cubic m of space and even then there's a daily charge per day if part isn't collected after 48hours)
Also, how much additional weight will the data centre floor need to support if there'll be reservoirs of dielectric fluid?
If we treat cooling as a utility, would vendors share these reservoirs of fluid?
I can see the Health & Safety people tut-tutting already at the very least.
If you flip the Arrhenius equation (The reaction rate doubles, when the temperature increases, for every 10 degrees Celsius*) on its head, you eventually end up with "every 10°C drop in temperature increases the life of electronics by a factor of two". Or for every 10 degrees Celsius lower the operating point of a Data Centre, half the number of parts will fail. Kind of makes you wonder if you fully dehumidified the air and dropped the operating temperature to say -30°C in a DC what the hardware failure rate would be like.
*The Arrhenius equation is heavily used by silicon chip makers, those who want to fully remove early failures, the "infant mortality" section of the "Bathtub curve". This is achieved by operating their chips inside ovens for a few days to weeks at high temperature to force chips to fail that would have failed in the first 6-12 months of operation to fault in a much shorter period (it is basically a form of super accelerated burn-in).
Yes actually a very good observation.
At the moment DC utilisation per unit space is limited by BTU cooling metrics "per tile" of space, right?
However, in reality once we reduce the cooling equation, won't they just pack the kit in, in increasing densities?
i.e. so called "efficiencies"?
With more kit again, it'll just drive up the failure rate again as a proportion of the assets. We only need a few pieces of kit dripping fluid for it just to be a hassle.
It reminds me of a time when the internet and computers were supposed to save us time and allow us to do the same work in less time, but instead we're just pushed to do more with less time (agile comes to mind).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
