High-performance computing (HPC) is already being adopted by companies looking to reach new levels of data analysis, and it’s essential to the future of AI and machine learning, but the first step to launching HPC is creating an environment capable of hosting it. HPC uses dense clusters of computer parts which need specialised infrastructure to cope with their power and cooling requirements.
The easiest way, for a lot of companies, to host their high-performance computing is to work with a data centre who provides high-quality IT infrastructure, but it’s important you work with a data centre capable of hosting HPC effectively, and with enough capacity to support your future system expansion.
Specialised infrastructure for high-performance computing
Due to HPC’s high density of parts, it requires advanced systems for powering and cooling, and this also means you need to find specialised HPC-ready data centres who are able to deliver this infrastructure.
Ordinary server racks in a data centre use around 3-7kW, but thanks to the increased number of CPUs and GPUs, an HPC rack can use upwards of 40kW in the same footprint. This is why specialised power infrastructure is required to deliver this high wattage and you need to find a data centre capable equipped with sufficient power delivery infrastructure.
The amount of heat generated by a rack is proportionate to the amount of power it uses, so HPC racks produce considerably more heat than an ordinary rack and need better cooling. Additionally, the high-density of clustered computer parts means traditional air-cooling systems aren’t effective at delivering cool air to all parts of the system. There is a wide range of specialised cooling infrastructure a data centre can install for HPC, anything from a rear-door cooling panel to immersion cooling.
At Redcentric, we use rear-door cooling in our data centres to effectively deliver cooling to high-performance computing racks. These use the same chilled water as the rest of the data centres cooling systems, but the cooling plate delivers it more directly to the servers (with specialised anti-leak technology deployed to protect your system).
Future proof your HPC system
HPC’s high requirements can make it difficult to plan for the future, but it’s essential you work with a data centre who can support you in the future, including providing you with the infrastructure you need to expand in the future.
With HPC’s high power and cooling demands, it’s common for this to be how a data centre reaches max capacity, before they fill up all of the physical racks on their data floor.
When you’re choosing a data centre, even if they have the capacity to host your high-performance computing system right now, make sure they have enough spare capacity so that you’ll be able to expand in the future. Otherwise your business growth could be stifled down the line if there’s a delay to expanding your high-performance computing system.
What else should you look out for?
An high-performance computing system is significantly more complicated than traditional computing, so it’s essential that it is hosted in an ideal environment for you to be able to run it at its maximum capacity and get the greatest return on your investment.
Data centres are often equipped with redundant equipment so that they can keep your system running even if there is equipment failure. Make sure this redundant equipment is also capable of supporting your HPC system, and not only suitable for traditional 3kW racks.
How to find an HPC-ready data centre
Not every data centre will be equipped for hosting your HPC system, so you might need to look further afield than you have previously. You also need to make sure that the data centre you end up working with is also capable of helping you expand your system in the future and is totally equipped for all of HPC’s requirements.