Liquid cooled data center solutions are vital to future efforts to balance High Performance Computing (HPC) and AI demands vs. the local electric power supply. The current electrical supply is not capable of meeting this demand long-term. One GPU consumes on average 54 kilowatt hours per day, the same equivalent of two residential homes. Now imagine 5,000 racks or 300,000 total GPUs in a data center and we’re up to 10,800,000 kilowatt hours – enough energy to power 400,000 homes or a city the size Austin, Texas.
If not cooled, these chips will never reach their maximum computational level, averaging under 75% capacity while taxing the legacy data center HVAC air condition equipment.
Liquid cooling solutions require multiple technologies, usually from different organizations, installed together sequentially, to create a sustainable data center system. The tricky part is to find a turnkey provider to plan, specify, coordinate, project manage, and install all of the technologies on-site for a seamless start up. Once installed, maintenance could become cumbersome on the end user. Liquid cooled solutions are increasing in adoption; but most data center technicians are not trained for proper liquid cooled hardware and component maintenance techniques.
Before starting your liquid cooling solutions evaluation, prepare by clearly defining your objectives and goals. Understanding your ideal business outcomes and organizational sustainability goals are critical steps to prepare for your solutions partner evaluation. As a second step, plan for a detailed audit of the data center project and your existing or targeted new equipment to help scope equipment and site requirements. Now you’re ready to present your objectives and find the right partner.
Your target partner should be a single-source provider that is not OEM biased and is well-versed in multiple liquid cooling technologies. This is the kind of open-minded provider that will tailor the optimal liquid cooling solution to meet your objectives. A single-source partner should be capable of leading initial specifications and scoping efforts, procure your target technology, deliver on-site project management and installation, install and commission your servers, and offer full lifecycle maintenance on your new system. If you’re evaluating multiple solutions providers, make sure to request a total cost of ownership summary from each of them for a true apples-to-apples comparison.
Liquid cooling technologies have a wide range of use cases and ranges of installation complexity. One of the most common technologies is RDHX “Rear Door Heat Exchanger”. This is the least intrusive liquid cooling solution and is capable of retrofitting to existing rack systems.
A second technology is Direct-To-Chip (DTC) cooling. DTC is more complicated to install but delivers higher cooling efficiency and enables higher rack density.
A third option is full liquid immersion. Full immersion is the most complex installation but provides significant advantages over RDHX and DTC. Full immersion benefits include maximum cooling efficiency and computational performance. Full liquid immersion also significantly reduces data center HVAC air conditioning requirements, dropping noise decibel levels for data center staff.
All three technologies have advantages and disadvantages to evaluate. Choosing an industry proven, single-sourced liquid cooling solutions provider is the first step in your journey to transform your data center.
