Making a Case for Green High-Performance Visualization Via Embedded Graphics Processors

This paper makes a case for using low-power embedded GPUs for the purpose of executing high-performance scientific visualization tasks. We compare the greenness (i.e., power, energy, and energy-delay product) of an embedded GPU with a CPU for commonly encountered visualization tasks using two real-world applications: (1) Modeling for Prediction Across Scale – Ocean (MPAS-O) and (2) Particular Ensembles (PE). Our preliminary results show that the low-power embedded GPU is capable of handling complex visualization tasks while consuming less than 50% of the energy consumed by a CPU server. In addition, we find that the embedded GPU outperforms the CPU with dynamic voltage-frequency scaling (DVFS) enabled in a majority of the cases.

[1]  Valerio Pascucci,et al.  Evaluation of In-Situ Analysis Strategies at Scale for Power Efficiency and Scalability , 2016, 2016 16th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid).

[2]  Wu-chun Feng,et al.  Towards efficient supercomputing: a quest for the right metric , 2005, 19th IEEE International Parallel and Distributed Processing Symposium.

[3]  Hank Childs,et al.  Exploring tradeoffs between power and performance for a scientific visualization algorithm , 2015, 2015 IEEE 5th Symposium on Large Data Analysis and Visualization (LDAV).

[4]  Wu-chun Feng,et al.  On the Greenness of In-Situ and Post-Processing Visualization Pipelines , 2015, 2015 IEEE International Parallel and Distributed Processing Symposium Workshop.

[5]  Scott Pakin,et al.  Exploring power behaviors and trade-offs of in-situ data analytics , 2013, 2013 SC - International Conference for High Performance Computing, Networking, Storage and Analysis (SC).

[6]  Sabela Ramos,et al.  Exploring energy-performance-quality tradeoffs for scientific workflows with in-situ data analyses , 2014, Computer Science - Research and Development.

[7]  Wu Feng,et al.  A Pipeline for Large Data Processing Using Regular Sampling for Unstructured Grids , 2017 .

[8]  Philip W. Jones,et al.  A multi-resolution approach to global ocean modeling , 2013 .

[9]  John Shalf,et al.  Exascale Computing Technology Challenges , 2010, VECPAR.

[10]  Manish Parashar,et al.  Exploring energy and performance behaviors of data-intensive scientific workflows on systems with deep memory hierarchies , 2013, 20th Annual International Conference on High Performance Computing.

[11]  Frank Mueller,et al.  A Power-Aware Cost Model for HPC Procurement , 2016, 2016 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW).

[12]  Hank Childs,et al.  PaViz: A Power-Adaptive Framework for Optimizing Visualization Performance , 2017, EGPGV@EuroVis.

[13]  Scott Pakin,et al.  Characterizing and Modeling Power and Energy for Extreme-Scale In-Situ Visualization , 2017, 2017 IEEE International Parallel and Distributed Processing Symposium (IPDPS).