enerGyPU monitor for workload characterization on Multi-GPU

enerGyPU main goal is to characterize the workload tasks and improve the balance between performance and energy-efficient. enerGyPU is a batch monitor formed by two levels:

1. The firts level automate the nvidia-smi queries to capture the power traces in runtime: in which the main launcher executes the energypu_record.sh in parallel with the scientific-application and write the GPU architectural factors.
2. The second level is a data visualization for analyzing the GPU architectural factors and model prediction system (EEA-Aware) for obtaining the optimal computational resources in a stactic time.

enerGyPU example on multi-GPU node:

The experimental procedures were executed with aset of test of HPL code variants using 6 GPUs on multi-GPU node. Work presented on Supercomputer Conference 2016; EEA-Aware for Large-Scale Scientific Applications on Heterogeneous Architectures.

Paper

If you use enerGyPU monitor, please cite this paper

@inproceedings{energypu,
  title={enerGyPU and enerGyPhi Monitor for Power Consumption and Performance Evaluation on Nvidia Tesla GPU and Intel Xeon Phi},
  author={John A. Garcia H., Esteban Hernandez B., Carlos E. Montenegro, Philippe O. Navaux, Carlos J. Barrios H.},
  booktitle = "{https://ieeexplore.ieee.org/document/7515761}",
  year={2016}
}

Project

enerGyPU Monitor was part of the master research work: Energy-Aware EEA for Large-Scale Scientific Applications on Heterogeneous Architectures (2016).

References

1. John A. G. Henao, Victor M. Abaunza, Philippe O. A. Navaux, Carlos J. B. Hernandez. eGPU for Monitoring Performance and Power Consumption on Multi-GPUs. (2015).

Help

Start an issue if you find a bug or would like to contribute! For other matters, you can contact @jagh.