Low-cost high-performance computing via consumer GPUs
Patrick Somaru, Iyad Obeid, Joseph Picone · 2016
Big data and machine learning are rapidly developing fields with evolving and increasingly diverse hardware requirements. The goal of this project was to demonstrate that an enterprise-ready, Warewulf-based HPC compute cluster could support heterogeneous hardware via the integration of a GPU compute server. The benefits of this were two-fold. First, the integration of the GPU compute server into the cluster would validate its ability to support heterogeneous hardware, demonstrating that our approach to HPC cluster management is an effective long-term strategy. Second, a GPU compute server would significantly reduce the runtime of experiments, increasing the rate at which research advancements are made.