Energy Efficient Computing

Aims

• At the Zettascale Lab, we are examining large-scale GPU and X86 systems to understand how power is currently used and distributed. Using a whole-system approach, by systematically analysing cause and effect, our aim is to gain insights into the best ways to balance the demands of speed and energy efficiency in zettascale class computing. We plan to develop energy efficient systems that make the best use of every joule that is spent. Not only will this maximise the value of IT investment by reducing operational cost, it will contribute to the problem of tackling environmental degradation.
• Currently, operations performed on high performance systems are not initially assessed for the amount of energy required to reach a solution. We are evolving a methodology to allow for an automated process that will estimate the amount of energy (and time) an operation takes to complete. The result will be dynamic and energy-aware system behaviour (e.g., a system that can react to the spot-pricing or availability of power). This could lead to a sector-wide shift whereby costs for operations are calculated on the basis of power consumption rather than time spent utilising cores. Software developers would then be incentivised to write code that made efficient use of the diverse components in an exascale system.
• The Zettascale Lab is designing training packages that will harness knowledge established about the design of energy efficient systems to promote coding for efficiency.
• In the domain of hardware, the Zettascale Lab is exploring innovative ways to reduce the power consumed in the process of heating and cooling systems. We are investigating the use of component-level liquid cooling.
• At the Zettascale Lab, we are designing tools to measure the power usage of our high performance systems at both the macro and micro level. This includes finding effective ways of visualising the components of the system, as well as monitoring temperatures and cooling methods at our data centre.