Intel Certifies Shell Lubricant for Cooling AI Data Centers (bloomberg.com) 36
Intel has certified Shell's lubricant-based method for cooling servers more efficiently within data centers used for AI. From a report: The announcement on Tuesday, which follows the chipmaker's two-year trial of the technology, offers a way to use less energy at AI facilities, which are booming and are expected to double their electricity demand globally by 2030, consuming as much power then as all of Japan today, according to the International Energy Agency.
So far, companies have largely used giant fans to reduce temperatures inside AI data centers, which generate more heat in order to run at a higher power. Increasingly, these fans consume electricity at a rate that rivals the computers themselves, something the facilities' operators would prefer to avoid, Intel Principal Engineer Samantha Yates said in an interview.
So far, companies have largely used giant fans to reduce temperatures inside AI data centers, which generate more heat in order to run at a higher power. Increasingly, these fans consume electricity at a rate that rivals the computers themselves, something the facilities' operators would prefer to avoid, Intel Principal Engineer Samantha Yates said in an interview.
Nothing new (Score:1)
Yeah, now it doesn't void the warranty, though. (Score:5, Insightful)
People have been dunking servers or other computing parts in oil for ages long time now an alternative to traditional cooling methods.
What you're missing is that they can do so without voiding your warranty. I did that LITERALLY 20 years ago with my first liquid cooled PC...back pre-kids when I had time for such shenanigans. It was FUN...but yeah, as soon as you remove the air cooler, you know you're not getting warranty support.
Also, regarding off the shelf stuff, you're taking a lot of risk. It doesn't take a lot of contamination to get a fluid to conduct enough electricity to short a circuit. I learned this the hard way on my first build 20 years ago. It was going great for a 3 months and then I did one upgrade...didn't notice a leak...so small it leaked less than a drop a day...after a few days, my video card died....barely any fluid on it...lesson learned...use non-ionic, treated fluid...basically specialty antifreeze. The water started out pure and distilled and non-condutive, but the waterblocks and fittings eventually reacted enough to change that...hence the need for the additives.
People have known for a long time that you can immerse a PC in mineral oil...but...are you SURE????...speakers will fail if there's an onboard one...magnetic hard disks will fail....for the bloggers 20+ years ago, it was fine...they didn't care, but it's one thing to boot a PC in an experimental configuration and post a blog post (or today it would be a YouTube video)...but....how does this thing hold up in a year? 5? 10? Is it robust enough to be moved? Are you sure the shit at CVS is pure?...I don't think salts should dissolve in mineral oil, but I'm not a chemist...does the mineral oil break down over time to form compounds that could conduct electricity just enough to short a circuit?
The value here is Intel is standing behind their products. They did the research and if they make updates, they'll ensure it works with this configuration. Shell is providing a specialty lubricant that is also certified to be pure enough...and they've done toxicity research...I don't know the details because the article is paywalled, so I don't know the markup, but I doubt it's that obscene.
It's one thing to put your old computer in a fishtank with CVS mineral oil for a fun science experiment...it's QUITE another to run a datacenter with machines that are intended to live 10 years or more and represent several millions of dollars of investment in hardware and presumably even more in lost revenue if all the machines were to suddenly fail. Presumably they have protocols and designs for maintenance as well...ways to make the whole replace a dead server process a lot less painful than a DIY solution.
Any sensible manager would want major companies certifying the process with all this money on the line. Anything that cools your servers more efficiently has the potential to greatly reduce costs, but this is a risky venture....lots of fun on my home PC, but I am not sure I have the appetite for risk to DIY a solution for thousands of servers...where failure would cost my job and quite a few others.
Re: (Score:2)
... does the mineral oil break down over time to form compounds that could conduct electricity just enough to short a circuit?
Or does it break down the glues and resins that hold bits and bobs together, resulting in package or PCB delamination and your nice pristine servers turning into piles of mush?
Re: (Score:2)
It's nice that you "think" that.
Re: (Score:2)
The good news is the lubricant is made from whale oil.
Re: (Score:2)
Re: (Score:2)
They probably also won't like that the Principal Engineer is named "Samantha".
Re: (Score:3)
the Principal Engineer is named "Samantha".
Is this one of those timelines where she didn't join the Air Force and become part of the Stargate Program?
Re: (Score:2)
Re: (Score:2)
Nothing about the end of the Energy Star program prevents manufacturers from continuing to sell energy-efficient products. They could even label them as 'energy-saving', refer to standards, even form an industry grip to are on those standards and such.
But blame the government for not incentivizing what should be a core goal of industry. Second only to profit...
Re: MAGA won't like this (Score:1)
Why shouldn't industry invest in electric utilities rather than produce energy efficient products? Are you confusing economic efficiency with engineering efficiency, as economists do implicitly?
Re: MAGA won't like this (Score:2)
All I'm trying to do is point out that manufacturers can continue to meet the Energy Star standards, if it's profitable to them. And consumers could ask about that...
Re: (Score:2)
its saving rich people money so its allowed.
Wouldn't moving all that fluid... (Score:1)
Cooling with a higher COP (Score:2)
Perhaps we need to do actual research on better cooling systems, perhaps hybrid phase change/desiccant systems which can have part of the cycle powered by solar (not PV, but heat, like an RV absorption fridge)?
What might be nice is a way to do cooling at the data center level, where a server can be plugged into recirculating coolant lines, with a valve system that, if it detects a leak would get the server to go to backup cooling, then evacuate the coolant from that server. This way, servers don't need to
Re: Cooling with a higher COP (Score:2)
"cooling at the data center level"
The idea sounds interesting. Like the way we deliver gas to our appliances. But would it be cheaper? To have a cooling system that's bespoke for server farms?
Cause on of the things that keeps data center prices manageable is leveraging economies of scale for the components whose manufacturing can be done alongside consumer products.
Re: (Score:2)
It can be done. One local university where I am has multiple chilling towers which replace A/C units for their entire campus. This means that every building only needs an air exchanger.
This would be in addition to other systems, for redundancy's sake, but a liquid subsystem is definitely possible. Open loop cooling is still not there yet, but closed loop cooling seems to be good enough for almost anything these days.
Cheaper? (Score:2)
I know you can get better cooling performance from liquid cooling, but is it actually cheaper to push liquid than push air?
Re: (Score:3)
since the liquid can move more heat per unit of volume than air, you don't need to move as much of it.
Re: Cheaper? (Score:2)
Sure, but I've never heard of someone putting together a liquid cooled rig to save on electricity.
Re: (Score:2)
And Intel isn't doing that either. The energy savings is just a side benefit to being able to increase thermal density within the data center by having more efficient means of moving that heat out of the data center.
Re: (Score:2)
My research group has two air-cooled Grace Hopper systems. This is a 72 core ARM processor, an H100 GPU and 480GB of memory all wedded on a single supermodule. The power dissipation i
Re: (Score:2)
I don't see how immersion cooling can compete with simple waterblock cooling. At datacenter level waterblock cooling adds plumbing as a new maintenance task, but if the water is kept deionized and the system is designed to gracefully handle leaks, a leak is not the end of the world.
The mess and sheer weight of submersion in transformer oil, which needs a refrigerant or water loop to get the heat away any way, seems far more trouble.
Additional information (Score:2)
I couldn't read the Bloomberg article because of the paywall. When I was looking for an alternative source I found this:
https://www.datacenterfrontier.com/cooling/article/55130995/turn-up-the-volume-data-center-liquid-immersion-cooling-advancements-so-far-in-2024
The article I just linked states that "such dielectrics often include mineral oil, fluorocarbon-based fluids, and deionized water". I'm guessing they must filter the coolant before re-circulating it, and/or replace it regularly. If they didn't do t
Re:Additional information (Score:4, Informative)
Use 100% synthetic motor oil. Install a pump, filter, low oil monitor, low oil pressure monitor (basically like a car engine). Use a heat exchanger, for example: a freon line in a coil, inside a larger tube carrying the oil.
For redundancy you can add a second pump and filter, possibly an alternator (not this one [wikipedia.org], this one [engineeringtoolbox.com].)
Re: (Score:2)
Interesting thought, but I'd also be concerned about petroleum products in contact with some of the plastics, and possibly rubber compounds as well, that might be present.
Re: (Score:2)
I think they're certifying this for use at the build stage. They would come stock with this setup. Other than the PCB, nothing strictly has to contain plastics. Though you do need insulation for wires. They would be able to pick the right materials up front.
Re: Additional information (Score:2)
I'm not sure that copy makes sense. But assuming it does, then it's because water is better at absorbing heat than oil. In that case the oil is provided as a lubricant and to help control some causes of corrosion. Think cutting fluid, which commonly uses water to remove heat, though there are multiple kinds of cutting fluid. (I've read that you can make great cutting fluid with bacon grease with 5% flower of sulfur, but I haven't tried it.)
Re: (Score:2)
Thanks. If they are using water for better thermal capacity, I guess it would have to be a pretty high percentage. As for the bacon grease, I'll try that the next time I have to tap a hole, just as an experiment.
I don't have the sulphur on hand though. The last time I did was over fifty years ago, when I used to make gunpowder and gunpowder-like mixes. Back then you could buy it, along with potassium nitrate, in drug stores. Powdered charcoal courtesy of Dad's barbecue and an electric drill. Came close to m
Re: (Score:2)
You are correct. If they were using pure water for coolant they would need to monitor and maintain quality -- but this is not a big issue, especially for a large expensive facility. But they probably aren't using water.
Optimization? (Score:2)
As an amateur radio operator, I pay some attention to how my equipment cools. And mobile radios use ducting as means to improve airflow and cooling.
I presume datacenters have evolved beyond 2000-era airflow and tunnels, and duct to the chassis. As well as use direct conversion power supplies to avoid the thermal losses and impact of ac-dc conversion.
Re: (Score:2)
As well as use direct conversion power supplies
An SMPS [wikipedia.org]? Yes. Since the 1981 IBM PC.
Re: Optimization? (Score:2)
Many switching regulators in there. I was thinking dc-dc. I've heard that dc our distribution is becoming the norm in data centers. Switching regulators probably there also, but not so bad perhaps.
A failure regardless, it still radiates heat. (Score:1)
Is there anything AI can't do? Oh yes, be efficient and help the environment.