Maximizing Power-Constrained Supercomputing Throughput

Maximizing supercomputing throughput within power and cooling limits is a key challenge for exascale systems, which are increasingly constrained by power rather than performance. Effective power and energy management is essential. Whereas power capping has been well-known to increase energy efficiency and reduce energy costs, power variability has emerged as an orthogonal driving force on cost through service pricing models and increase component wear out. We present a performance-power-efficiency model that ties application performance, empirical power usage, power variability, and energy efficiency in a single methodology to enable optimize HPC system operation. We demonstrate the ability of our methodology to understand performance and energy efficiency through power capping on NVIDIA GPUs using seven workloads and three microbenchmarks. We show that uniform throughput power capping can potentially improve power-constrained system throughput by 1.8x and 2.5x based on capped node power and peak node power, outperforming uniform power capping by 20%.

---

Contributors:

• Nan Ding
• Oscar Antepara
• Brian Austin
• Leonid Oliker
• Samuel Williams