As the power consumption of HPC systems will be a primary constraint for exascale computing, a main objective in HPC communities is recently becoming to maximize power efficiency (i.e., performance per watt) rather than performance. Although programmers have spent a considerable effort to improve performance by tuning HPC programs at a code level, tunings for improving power efficiency is now required. In this work, we select two representative HPC programs (Graph500 and SDPARA) and evaluate how traditional code-level performance tunings applied to these programs affect power efficiency. We also investigate the impacts of the tunings on power efficiency at various operating frequencies of CPUs and/or GPUs. The results show that the tunings significantly improve power efficiency, and different types of tunings exhibit different trends in power efficiency by varying CPU frequency. Finally, the scalability and power efficiency of state-of-the-art Graph500 implementations are explored on both a single-node platform and a 960-node supercomputer. With their high scalability, they achieve 27.43 MTEPS/Watt with 129.76 GTEPS on the single-node system and 4.39 MTEPS/Watt with 1,085.24 GTEPS on the supercomputer.

A key challenge in next-generation supercomputing is to effectively schedule limited power resources. Modern processors suffer from increasingly large power variations due to the chip manufacturing process. These variations lead to power inhomogeneity in current systems and manifest into performance inhomogeneity in power constrained environments, drastically limiting supercomputing performance. We present a first-of-its-kind study on manufacturing variability on four production HPC systems spanning four microarchitectures, analyze its impact on HPC applications, and propose a novel variation-aware power budgeting scheme to maximize effective application performance. Our low-cost and scalable budgeting algorithm strives to achieve performance homogeneity under a power constraint by deriving application-specific, module-level power allocations. Experimental results using a 1,920 socket system show up to 5.4X speedup, with an average speedup of 1.8X across all benchmarks when compared to a variation-unaware power allocation scheme.

Recently to achieve the Exa-flops next generation computer system, the power consumption becomes the important issue. On the other hand, the power consumption character of application program is not so considered now. In this study we examine the power character of our Magneto hydrodynamic (MHD) simulation code for the global magnetosphere to evaluate the power consumption behavior of the simulation code under the CPU power capping on the parallel computer system. As a result, it is confirmed that there are different power consumption parts in the MHD simulation code, which the execution performance decreases or does not change under the CPU power capping. This indicates the capability of performance optimization with the power capping.

To predict application performance on an HPC system is an important technology for designing the computing system and developing applications. However, accurate prediction is a challenge, particularly, in the case of a future coming system with higher performance. In this paper, we present a new method for predicting application performance on HPC systems. This method combines modeling of sequential performance on a single processor and macro-level simulations of applications for parallel performance on the entire system. In the simulation, the execution flow is traced but kernel computations are omitted for reducing the execution time. Validation on a real terascale system showed that the predicted and measured performance agreed within 10% to 20 %. We employed the method in designing a hypothetical petascale system of 32768 SIMD-extended processor cores. For predicting application performance on the petascale system, the macro-level simulation required several hours.

A full electron calculation for the photosynthetic reaction center of Rhodopseudomonas viridis was performed by using the fragment molecular orbital (FMO) method on a massive cluster computer. The target system contains 20,581 atoms and 77,754 electrons, which was divided into 1,398 fragments. According to the FMO prescription, the calculations of the fragments and pairs of the fragments were conducted to obtain the electronic state of the system. The calculation at RHF/6-31G* level of theory took 72.5 hours with 600 CPUs. The CPUs were grouped into several workers, to which the calculations of the fragments were dispatched. An uneven CPU grouping, where two types of workers are generated, was shown to be efficient.