With only one, perhaps two percent of the larger data center cooling market share, immersion and direct liquid cooling are still marginal technologies. The reasons are as nuanced as they are numerous, especially for immersion.
Although it has been around for a number of years, the reputation for maintenance, leaks and spills has quickly overtaken the benefits, which include nearly 100% thermal efficiency. These potential drawbacks, coupled with concerns about floor space and facility integration (supporting the weight of immersion tanks, for example) as well as how the introduction of liquids might alter support contracts with system providers are further complications.
But a day will come, most likely this decade, when it might be the only option, at least for high density servers in data centers supporting supercomputing applications or AI training, among other workloads. demanding work. Some HPC centers have taken the route of immersion, but again, these are exceptions.
Lucas Beran, senior analyst for Dell’Oro’s data center infrastructure group, says that today, less than 1% of servers worldwide are cooled by direct liquid cooling or by immersion. “When I talk to engineers, they say they like the idea of immersion and liquid. But operationally, the barrier is the human element. Data center owners and operators don’t want to add liquids and oils to the data center environment, they worry about damage and destroyed equipment.
While the “human element” maintains the adoption of remote immersion technology, Beran says the resistance cannot continue any longer. “I don’t think we’ll get to a larger immersion until we need to get there, but we’re starting to be there.” Densities are increasing rapidly and we are rapidly approaching the tipping point.
The movement towards immersion will begin in environments of extreme density. “Direct cooling of a processor or GPU directly on the chip is not 100% heat capture. If you have a rack that generates 40kW of heat, you will still have around 4-8kW of heat escaping, meaning you will need hybrid cooling via attaching a back door unit or an air handling system or other form of air cooling. to deal with this heat. Going forward, with extreme HPC rack densities of up to 200 kW if you only capture 80% of that heat, there is still a lot to cool, ”says Beran.
In other words, now is the time to at least start looking at what immersion cooling might require. Although implementation can take years, it is essential to design facilities and even support contracts with system vendors in advance. Air cooling might not go away anytime soon, but in some cases it won’t be enough.
“We are really only at the start of a decade-long transition from air-based cooling to liquid-based cooling,” says Beran. “The perimeter is a legacy technology, the rack and row or back door swaps, especially for high density sections or hot spots, are a middle part between this last frontier of thermal management, which is cooling. by immersion. “
If we are facing an immersion cooling future, which vendors to watch out for now, how are they different, and how much room for innovation do startups have to tackle market share? And could all the current dynamics of a standalone business (no matter how small) be turned upside down if an HPE or Dell, for example, decides to integrate it into their offerings for some of these higher density environments?
Beran says Green Revolution Cooling (GRC) (which is installed at the supercomputing site, TACC, for example) is now the leader in immersion. Asperitas and Submer are honorable mentions with still others including Isotope garnering some opinion. “There are other startups in the space with some proofs of concept now and the market could change quickly, but I’m confident in GRC and Asperitas – dethroning either would be difficult, even if it is. still early.”
Differentiation for these companies and future companies is now focused on adopting a one- or two-phase immersion approach. In single phase, the liquid enters the tank, captures the heat, then is pumped through a heat exchanger before returning there. For the two-phase, once the liquid in the tank reaches a certain temperature, it vaporizes, rises, is redirected for cooling elsewhere, and then reform back into liquid. The latter has a higher CAPEX and although it offers improved cooling, it is so far “marginal” according to Beran.
The other area of differentiation lies in the technical fluid which is in the reservoir. Asperitas is working with Shell to refine its medium, 3M is working on its own immersion fluid. But at the moment, it’s hard to say what a big difference the fluid improvements will make. In short, they can allow startups to differentiate themselves technologically, but for something that’s already hard to sell, offering one-phase incremental improvement might not resonate and taking a two-phase approach might add even more. complications and more. -cost before. So far, by the way, the major immersion companies mentioned are all single phase. Moreover, even though its GRC material, it is a good explanation on the differences between single phase and two phase.
Perhaps the only differentiation that could make practical sense at this early stage would be if immersion became a core offering with full systems-level support from the major OEMs providing the bulk of the data centers for these. high density data centers in HPC or AI training. For example, if HPE / Cray had an immersion option guaranteeing these systems: servers, storage, networks and immersion tanks. Currently, a concern is that the introduction of immersion or direct liquid could invalidate a support contract. If a major OEM were to buy a GRC, for example, and cool a specific line for those use cases, that would be a different set of considerations.
It all comes back to the questions that preoccupy data center operators: what damage and what dangers to equipment do these tanks represent? What about safety in the event of a slip? What about damage and maintenance? There is a chicken / egg scenario here. More centers will need to embrace immersion cooling and openly share their challenges so that others can assess the risks. But no one wants to go first.
Besides these most talked about reasons, there are other practical issues that prevent submersion, including on the facility side. These reservoirs aren’t small – they take up a fair amount of data center floor space and require some engineering to support. In other words, implementation requires planning and if everyone is waiting for someone else to start, the rollout of more immersion examples is further delayed.
And speaking of delays, while submersion might be the only way to cool high-performance hardware over the next decade, its growth has been further hampered by the pandemic. Beran says it was a “high end” purchase from the start that required ground boots along the way. Even barring other stops, 2020-2021 could have pushed the entry of immersion into more mainstream cooling even further.
The final question is whether there is an opportunity for a startup hoping to gain a foothold in what appears to be one of the only options for near 100% heat capture. The answer depends on who is acquired by a major OEM with a reasonable base in the HPC / AI server markets (HPE, Dell, Lenovo and to be polite, maybe IBM). Internationally, companies like Fujitsu, a pioneer in immersion cooling for supercomputers in particular, have already invested, although we still don’t see many large publicly traded systems that incorporate immersion.
The equipment manufacturers have in particular forged partnerships with GRC (Dell and HPE, the latter also has a partnership with Iceotope but what the large centers want is integrated support for systems with this unique and particularly risky technology, it is not ruled out that one will be purchased, forcing a new competitive landscape in a game that is still far too early to call.