Degree

In the cooperation project “Degree“, using the example of the High Performance Computing Center Stuttgart (HLRS), a dynamic control of the cooling circuit temperatures shall be investigated to increase the efficiency of the cooling system.

The key interfaces of the internet are computing centers. Beside the servers, the operation of the cooling systems is the second large consumer of electrical energy in computing centers and currently often responsible for a third out of the total energy demand. The most energy-efficient way of cooling is “free cooling”: waste heat is delivered to the ambient air via wet cooling towers or dry cooling towers. This will only work if the temperature level of the waste heat is above the outdoor temperature. Higher operational temperatures are therefore pursued. However, the increase in the temperature of the coolant also results in negative effects. Due to higher operational temperatures, numerous components within the computing centers consume more power, especially semiconductors such as CPUs and main memories. In addition, also the computing power and the service lives of the IT components can decrease.

The “Degree“ research project addresses this issue between lowest possible operational temperatures and a maximum of free cooling (see Figure 1). This shall be achieved by the development of a method to implement dynamic cooling circuit supply temperatures. The supply temperatures are increased just to a level that free cooling is always possible and that unnecessarily high supply temperatures are prevented. Due to the temperature dependencies of the different components, the entire system is considered.

The achievable improvements are quantified and evaluated as numerical simulations: energetically, ecologically, economically, and related to computing power. The findings are realized and tested at HLRS. Figure 2 shows the simplified cooling system of HLRS. With a computing power of 26 PetaFLOPS, the HAWK supercomputer of HLRS is one of the fastest supercomputers in Europe.

The objective of the research project is to determine the achievable improvements by the operating mode with dynamic cooling circuit temperatures and to develop a method for implementation. An optimized control strategy of the cooling system is included. For the transferability to other computing centers, a manual will be created with specific guidance for the implementation of the developed method.

6/2021 – 6/2023