enerGyPU monitor for workload characterization on Multi-GPU

Role: HPC Systems Architect
Start Date: Jun 1, 2014
End Date: May 31, 2016
Status: Completed
Institution: High Performance and Scientific Computing Unit, Industrial University of Santander, Colombia.
Funding Institution:
Website: https://github.com/jagh/enerGyPU?tab=readme-ov-file

Summary

enerGyPU: Workload Characterization on Multi-GPU Systems

enerGyPU aims to balance performance and energy efficiency by characterizing workload tasks. It consists of two levels:

Monitoring: Automates nvidia-smi queries to capture power traces at runtime. The main launcher runs energypu_record.sh alongside the scientific application to record GPU architectural factors.
Analysis and Prediction: Visualizes GPU data and uses the EEA-Aware model to predict optimal computational resources.

Nvidia GPU Technology Conference (GTC) 2016. enerGyPU for Monitoring Performance and Power Consumption on Multi-GPUs. San Jose, USA.
IEEE/ACM 16th International Symposium on Cluster, Cloud and Grid Computing, CCGrid 2016. enerGyPU and enerGyPhi Monitor for Power Consumption and Performance Evaluation on Nvidia Tesla GPU and Intel Xeon Phi. Cartagena, Colombia.
Supercomputing Conference SC16. Energy-awareness to Accelerate Large-scale Scientific Applications in Heterogeneous Architectures. The International Conference for High Performance Computing, Networking, Storage, and Analysis. Salt Lake City, Utah, USA.