Liquid immersion cooling is increasing in popularity due to the efficiencies it offers organisations when considered against traditional cooling methods. Diarmuid Daltún, CCO, Submer Technologies, tells us more about the technology and the benefits it offers.
What challenges are data centre operators encountering when it comes to power and cooling?
The solutions data centre operators have in place today are very inefficient – they are using way more electricity and energy than they should be. They are also coming to the limit of how much they can put in the space of one rack space or one cabinet. With air, they are able to cool up to 30kW, but only by using a mix of methods, such as forced air and rear door cooling.
What is liquid immersion cooling and how is it addressing these challenges?
LIC is a method of cooling hardware by having it immersed directly in the primary coolant. In Submer’s case, the coolant is circulated actively around the hardware, taking the heat away from every component (any component which consumes electricity, generates heat). The heat is then passed out from the coolant to a secondary cooling loop which is typically a closed water system connected to a dry cooler or a method of heat reuse (e.g. district heating or building heating).
What kind of data centre operators are utilising this method?
The initial interest is coming from HPC, since they are facing with the cooling challenge on a daily basis, due to their dense workload. The LIC method is also thought for any application which is used for research purposes which is likely to put strain on the limits of air cooling (e.g. AI, baking, enterprise, hyperscaler, etc.).
How scalable are liquid immersion cooling solutions?
Speaking of Submer, our solution is very easily and modularly scalable. Each Submer’s tank (SmartPod) can dissipate >50kW and can be deployed anywhere, without any preparation of the site. You can place them next to each other or back to back (according to your facility needs) since they occupy very little space compared to air-cooled racks.
How does immersion cooling differ from water cooling?
Water cooling is a very broad topic. Direct liquid cooling (water piped directly to some components, such as GPU, CPU, etc.) for example, requires a specific type of hardware and another method of cooling inside the building. Direct liquid cooling deals typically with 70% of the heat. Immersion cooling deals with the entire heat footprint, doesn’t care about the shape of the equipment and doesn’t need to be specifically prepared (it only requires a hardware with no moving parts, such as fans).
What are the varying types of immersion cooling?
Passive immersion cooling relies on convection alone to move the fluid around for the heat to be removed (not suitable for high densities). In Submer’s experience, the liquid must be actively circulated in order to ensure that there is a low delta between the coolant temperature and the temperature of the component to be cooled.
There are solutions that share a CDU (Cooling Distribution Unit) with a number of tanks. The challenge in this case is that the coolant (typically a mineral oil) leaves the CDU and this increases the chances of leaks. In this case also, the CDU works according to the heat generated in all the tanks, not adjusting to the workload of each pod.
Then we have single-phase and two-phase immersion cooling. In the latter, the coolant changes its states from liquid to gas into liquid again. The liquid is boiling directly on the components, generating though challenges such as micro-cavitation and evaporation of the coolant itself (that is extremely expensive). In the single-phase solution, as in Submer’s, the liquid never changes its states and it doesn’t leave the tank.
What kind of efficiencies could a business expect to achieve by utilising these solutions?
The typical PUE (Power Usage Effectiveness) for a data centre enterprise is 1.89 (it depends on many factors). This means that for 1kW of energy used to power a server, 0.89kW are used to power the rest of the picture. If you check the Submer’s SmartPUE calculator, you can notice that a PUE of 1.89 actually refers to a real PUE of 2.36. In this scenario, more than half of the electricity consumed is used for cooling (air condition and fans on servers).
With LIC, you manage to get a PUE <1.03: this means that for every 1kW of real IT (i.e. servers with NO fans) being powered, you’ll just need to add 0.03kW for primary and secondary cooling together. So, essentially you need less than half of your electricity compared to a typical, traditional air-cooled data centre.
How easily could an organisation transition to this type of solution?
An existing installation can adopt an immersion cooling solution by modifying the existing servers (by removing the fans) to submerge them. You should also invest in secondary cooling infrastructure. Still in such a scenario, the energy savings are so significant that the ROI is less than one year. Implementing immersion cooling solutions is definitely much easier when it comes to completely new building (a typical new data centre is around US$5 to US$10 million per MW, with immersion cooling it is US$3 million per MW).
In summary, what are the main benefits of liquid immersion cooling?
Density (85% more denser than with air cooling), efficiency (50% savings on electricity), you can deploy it anywhere (reducing building costs), it is a silent cleantech and it reduces latency and increases deployment speed.