One of the most significant limitations of high performance computing is energy. HPC solutions typically utilize a tremendous amount of energy, to a point that is often difficult to provide or justify. As more organizations in both the private and public sector seek to leverage HPC tools for a growing variety of purposes, these energy issues are likely to become increasingly important.
Recently, Natalie Bates spoke to Scientific Computing to discuss this issue. Bates is the chair of the Energy Efficient High Performance Computing Work Group (EE HPC WG), an organization dedicated to encouraging the implementation of more energy-efficient HPC designs. As Bates explained, energy efficiency is still a major challenge for HPC, but progress is being made.
According to Bates, one of the most significant obstacles preventing more energy-efficient HPC is the limited scalability of silicon-based technologies, which are used to fabricate HPC system parts. She noted that there are a number of alternative computing strategies that could potentially deliver similar results while avoiding silicon components, including quantum computing, biomolecular computing and optical computing, but that none of these has progressed beyond the early research stages yet.
"[A]lthough they are important to pursue, none of them can be counted on to help with creating more energy-efficient HPC for the near future," said Bates, the news source reported.
The most realistic alternative on the horizon, she explained, is what Google's Barroso and Hozle termed "energy proportional computing." This is not so much an alternative to current HPC as a new way of optimizing energy use by ensuring that resources do not consume energy when not performing useful work.
Another issue preventing more energy-efficient HPC solutions is the lack of standardization and metrics, Bates told the news source.
"Depending on the target group (e.g., application developers, system integrators, HPC data center administrators), the expectations and goals differ – sometimes these goals and optimizations conflict with each other," she said, Scientific Computing reported. "Should we care about FLOPS per Watt, or cost per Watt, or Energy-Delay products, or exceeding a power budget, or science accomplished per watt or utilizing allocated power well?"
However, despite these challenges, the EE HPC WG has made significant progress toward improving energy efficiency in the HPC sector, Bates asserted. She explained that the group is comprised of more than 450 members from 20 countries, with the majority of active participants based in large supercomputing centers in the United States Department of Energy Laboratories or European counterparts.
"This collective voice provides a strong influence to encourage system integrators, standards bodies and other organizations to actively participate in the drive for energy efficiency measures and design," she said, the news source reported.
Specifically, Bates noted that the EE HPC WG has developed guidelines for liquid cooling inlet temperatures, improved power measurement methodology for use while running workloads and benchmarks, revamped the data center infrastructure energy efficiency metric TUE and more.
Excessive energy use is not the only challenge facing organizations as they increase their reliance on HPC solutions in the coming years. As these systems and the codes used on them become more complex, firms need to be wary of the growing risk posed by bugs. These flaws are difficult to spot and can seriously undermine the HPC system as a whole.
That is why firms utilizing HPC resources need to invest in high-quality debugging solutions. These tools should feature advanced visualization capabilities to help personnel better troubleshoot any problems that arise when leveraging HPC assets.