What question did this study set out to answer?

To introduce RAJA, a C++ abstraction library that enables performance portability across diverse computing architectures from a single-source codebase.

November 1, 2019Open Access

RAJA: Portable Performance for Large-Scale Scientific Applications

View Full Paper

Ask AI

Bookmark

To introduce RAJA, a C++ abstraction library that enables performance portability across diverse computing architectures from a single-source codebase.
Integrated the RAJA portability layer into three large-scale production scientific codes at Lawrence Livermore National Laboratory.
Benchmarked application runtimes across heterogeneous hardware configurations, comparing GPU-only node performance directly against CPU-only node performance.
Achieved a 17× speedup on GPU-only over CPU-only nodes in the first production application.
Demonstrated 13× and 12× speedups on GPU-only over CPU-only nodes in the second and third production applications, respectively, while preserving a single-source codebase.

Discussion

No takes yet. Share an insight, caveat, or question.

Implication

Computational benchmarking demonstrates 12× to 17× GPU speedups across production scientific codes, indicating effective performance portability using single-source C++ software.

Cite This Study

Beckingsale et al. (2019) studied this question.

synapsesocial.com/papers/6a12ed5306ed52b5c2c0c871 https://doi.org/10.1109/p3hpc49587.2019.00012

View Full Paper

Ask AI

Bookmark

Implication

Computational benchmarking demonstrates 12× to 17× GPU speedups across production scientific codes, indicating effective performance portability using single-source C++ software.

RAJA: Portable Performance for Large-Scale Scientific Applications

Key Points

Discussion

Implication

Cite This Study

RAJA: Portable Performance for Large-Scale Scientific Applications

Key Points

Discussion

Implication

Cite This Study

Also Consider

Also Consider