As energy efficiency becomes a growing concern in high-performance computing, the POP3 Centre of Excellence portfolio of tools has been extended with tools to help developers and users understand and optimise the energy usage of their parallel applications. These are the MERIC runtime system and the RADAR visualizer from the MERIC energy efficiency HPC software suite for datacenter power monitoring, budgeting, and power management, developed at the IT4Innovations National Supercomputing Center starting already in 2015. These can be used to analyse a parallel application's energy consumption and energy-efficiency-related metrics.
MERIC is a tool designed to evaluate and optimize the energy efficiency of HPC applications. It enables the dynamic adjustment of hardware and system-software power management knobs during application execution, allowing users to achieve energy savings with a user-defined performance trade-off. MERIC can be used as a command-line utility or as a library to instrument C, C++, or Fortran applications using OpenMP, MPI, CUDA, or HIP for parallelization. Beyond tuning, it collects and reports various power-related hardware metrics to give insight into how computing resources are utilized throughout runtime. Depending on the hardware platform, this may include total energy consumption, power usage over time, energy efficiency indicators, power capping activity, thermal conditions, or CPU frequency data. These metrics tell how efficiently a hardware platform is used and help identify regions of the application where MERIC can save power without impacting performance.
RADAR visualizer is a visual analysis tool that complements the runtime system by providing a user-friendly graphical interface for interpreting the collected metrics. It helps users quickly understand the application’s behaviour during runtime and supports the broader POP methodology for energy efficiency assessments. By turning raw measurement data into accessible visualizations, RADAR makes it easier for developers and users to spot inefficiencies and identify opportunities for tuning and optimization.
These tools are already regularly used on production systems such as MACC's Deucalion or IT4Innovations' machines, particularly in the context of collaborative work with EuroHPC projects such as SCALABLE, MaX, SPACE, and the POP Centres of Excellence. The MERIC runtime system is currently being deployed on EuroHPC systems and will soon be available as a public software module. This will allow users to easily begin measuring the energy consumption of their applications in their own HPC environments.
Looking for hands-on support? Through the POP3 Centre of Excellence, you can request a second-level service where energy experts will perform an in-depth analysis of your application using MERIC – at no cost. You’ll receive a detailed report identifying inefficiencies and suggesting energy-saving strategies specific to your code and hardware target.it4i meric