Asymmetric Computation for Speculative Heterogeneous HPC

Lorenzo Altamura, Stefano Conoci, and Alessandro Pellegrini


Published in: 15th International Conference on High Performance and Embedded Architecture and Compilation Workshops
pdf Download PDF

Abstract:
HPC applications on future exascale systems will demand for runtime environments able to transparently manage the complexity of the underlying heterogeneous hardware. In this abstract, we discuss a computation model for speculative HPC applications, able to deliver non-minimal performance increase and significant energy savings. This model can be easily adapted to multiple heterogeneous hardware families with minor effort, and it can autonomically and promptly reassign units of work to different hardware classes. Our design jointly targets performance and energy efficiency. We also provide a preliminary experimental evaluation of our design.

BibTeX Entry:

@inproceedings{Alt20,
author = {Altamura, Lorenzo and Conoci, Stefano and Pellegrini, Alessandro},
booktitle = {15th International Conference on High Performance and Embedded Architecture and Compilation Workshops},
title = {Asymmetric Computation for Speculative Heterogeneous HPC},
year = {2020},
month = jan,
series = {HiPEAC},
location = {Bologna, Italy}
}