GPU Variability

Characterizing Variability in Large-scale, Accelerator-rich Systems

In this work, we characterized GPU variability in GPU supercomputers such as ORNL Summit and analyzed how it is affected by cluster attributes like size, cooling, and GPU vendors. We also profiled different applications from various domains (image processing, machine translation, molecular dynamics, graph analytics) on the same cluster to evaluate application-dependence. I co-authored and also presented this work at Supercomputing (SC) 2022 in Dallas, TX (Sinha et al., 2022).

References

2022

Not All GPUs Are Created Equal: Characterizing Variability in Large-Scale, Accelerator-Rich Systems

Prasoon Sinha, Akhil Guliani, Rutwik Jain, and 3 more authors

In SC22: International Conference for High Performance Computing, Networking, Storage and Analysis , Nov 2022

DOI Bib

@inproceedings{Sinha-SC22-GPUVar,
  author = {Sinha, Prasoon and Guliani, Akhil and Jain, Rutwik and Tran, Brandon and Sinclair, Matthew D. and Venkataraman, Shivaram},
  booktitle = { SC22: International Conference for High Performance Computing, Networking, Storage and Analysis },
  title = {{ Not All GPUs Are Created Equal: Characterizing Variability in Large-Scale, Accelerator-Rich Systems }},
  year = {2022},
  volume = {},
  issn = {},
  pages = {1-15},
  doi = {10.1109/SC41404.2022.00070},
  url = {https://doi.ieeecomputersociety.org/10.1109/SC41404.2022.00070},
  publisher = {IEEE Computer Society},
  address = {Los Alamitos, CA, USA},
  bibtex_show = true,
  selected = true,
  month = nov
}