The increasing power demands of HPC systems are putting a strain on the environment. For example, when it operates continuously, Frontier, the world's top-ranked supercomputer, requires around 20 megawatts of power. This is equivalent to the energy consumption of 15,000 homes. To minimize the environmental impact of HPC, there are two primary approaches. One is to improve hardware's energy efficiency, while the other focuses on developing energy-aware algorithms.
While hardware has made strides in recent years, its progress rate is slowing due to the physical limits of thermodynamics. A more promising way to reduce the environmental impact of HPC is to develop more energy-aware algorithms. Enhanced energy efficiency can be attained through strategic optimizations of computational loads, such as reducing the precision of less accuracy-sensitive kernels. One straightforward approach involves reorganizing parallel computations to enhance the efficiency of the application's computationally intensive segments, thereby minimizing the execution time at maximal power consumption. This allows for the utilization of low-precision energy-efficient processing units, resulting in energy conservation without significantly impacting the accuracy of the results.
Finally, a new method is to offload the preprocessing to the edge, which entails conducting some computation on less powerful but more energy-efficient devices closer to the data source. This can curtail the volume of data that must be transmitted to the HPC system and processed.
This Research Topic aims to introduce novel, promising research trends on algorithmic techniques for sustainable HPC.
The authors are invited to submit their studies on the feasibility of developing fast and energy-efficient software for HPC. The specific areas may include, but are not limited to:
• Identify critical algorithmic challenges in sustainable HPC (e.g., compression and transmission, load balancing, communication-avoiding techniques).
• Investigate suitable parallelization techniques for energy-safe algorithm implementation.
• Consider the most innovative HPC infrastructures and how algorithms can exploit them to save energy.
• Present new methods and algorithms for solving large-scale dimension problems more energy-efficiently than the existing ones.
• Evaluate the energetic performance of new methods and algorithms on real-world applications.
• Compare the energy performance of commonly used software in an HPC environment to select the most suitable one.
Potential authors are invited to submit Original Research, Systematic Reviews, Methods, and Technology and Code papers.
Keywords:
Sustainable HPC, Algorithmic techniques, Energy efficiency, Low-precision algorithms, Edge offloading
Important Note:
All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.
The increasing power demands of HPC systems are putting a strain on the environment. For example, when it operates continuously, Frontier, the world's top-ranked supercomputer, requires around 20 megawatts of power. This is equivalent to the energy consumption of 15,000 homes. To minimize the environmental impact of HPC, there are two primary approaches. One is to improve hardware's energy efficiency, while the other focuses on developing energy-aware algorithms.
While hardware has made strides in recent years, its progress rate is slowing due to the physical limits of thermodynamics. A more promising way to reduce the environmental impact of HPC is to develop more energy-aware algorithms. Enhanced energy efficiency can be attained through strategic optimizations of computational loads, such as reducing the precision of less accuracy-sensitive kernels. One straightforward approach involves reorganizing parallel computations to enhance the efficiency of the application's computationally intensive segments, thereby minimizing the execution time at maximal power consumption. This allows for the utilization of low-precision energy-efficient processing units, resulting in energy conservation without significantly impacting the accuracy of the results.
Finally, a new method is to offload the preprocessing to the edge, which entails conducting some computation on less powerful but more energy-efficient devices closer to the data source. This can curtail the volume of data that must be transmitted to the HPC system and processed.
This Research Topic aims to introduce novel, promising research trends on algorithmic techniques for sustainable HPC.
The authors are invited to submit their studies on the feasibility of developing fast and energy-efficient software for HPC. The specific areas may include, but are not limited to:
• Identify critical algorithmic challenges in sustainable HPC (e.g., compression and transmission, load balancing, communication-avoiding techniques).
• Investigate suitable parallelization techniques for energy-safe algorithm implementation.
• Consider the most innovative HPC infrastructures and how algorithms can exploit them to save energy.
• Present new methods and algorithms for solving large-scale dimension problems more energy-efficiently than the existing ones.
• Evaluate the energetic performance of new methods and algorithms on real-world applications.
• Compare the energy performance of commonly used software in an HPC environment to select the most suitable one.
Potential authors are invited to submit Original Research, Systematic Reviews, Methods, and Technology and Code papers.
Keywords:
Sustainable HPC, Algorithmic techniques, Energy efficiency, Low-precision algorithms, Edge offloading
Important Note:
All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.