Energy consumption is one of the biggest challenges when designing and building more powerful supercomputers. Also, a large proportion of the energy used by different computer components is converted to heat, and effort and possibly energy are needed to cool down the supercomputer. Therefore, the main reason for moving into massively parallel computing instead of making single CPUs more powerful was energy consumption. Also, GPUs are used in supercomputers because they can provide better performance per energy for certain workloads.
The second fastets supercomputer in the world in June 2023, the Japanese Fugaku, has a power consumption of 28 MW, and even though it is within the ten most energy-efficient computers in the world, its energy consumption is comparable to 10,000 average Finnish households. As the importance of energy efficiency is rapidly increasing, supercomputers are measured not only by their raw computing power but also by their energy efficiency on the Green 500 list. In the battle against global climate change, supercomputers are a double-edged sword. On the one hand, they help understand and fight climate change, but on the other hand, they use large amounts of energy.
Due to the dense packing of supercomputers, cooling can be more challenging than in domestic devices. However, proper climate can help in cooling, and modern data centers also have lots of cooling infrastructure. The two most common approaches are air cooling, where a steady airflow is run through the supercomputer, and liquid cooling, where water or other suitable liquid is circulated through the system carrying the heat away. In addition, waste heat can often be used for warming up premises close to the data center and, in best cases, even nearby municipalities.
Air cooling is relatively simple but not very efficient. Modern densely-packed supercomputers, especially GPU-based, require liquid cooling to keep them from melting down. As a result, a data center hosting supercomputers needs a massive and complicated cooling infrastructure with piping, pumps, heat exchangers, etc.
In CSC’s Kajaani data center, LUMI's waste head contributes to the local municipal household heating.