SC19 Proceedings

The International Conference for High Performance Computing, Networking, Storage, and Analysis

Poster 99: Eithne: A Framework for Benchmarking Micro-Core Accelerators

Authors: Maurice C. Jamieson (Edinburgh Parallel Computing Centre, University of Edinburgh), Nick Brown (Edinburgh Parallel Computing Centre, University of Edinburgh)

Abstract: Running existing HPC benchmarks as-is on micro-core architectures is at best difficult and most often impossible as they have a number of architectural features that makes them significantly different from traditional CPUs : tiny amounts on-chip RAM (c. 32KB), low-level knowledge specific to each device (including the host / device communications interface), limited communications bandwidth and complex or no device debugging environment. In order to compare and contrast different the micro-core architectures, a benchmark framework is required to abstract much of this complexity.

The modular Eithne framework supports the comparison of a number of micro-core architectures. The framework separates the actual benchmark from the details of how this is executed on the different technologies. The framework was evaluated by running the LINPACK benchmark on the Adapteva Epiphany, PicoRV32 and VectorBlox Orca RISC-V soft-cores, NXP RV32M1, ARM Cortex-A9, and Xilinx MicroBlaze soft-core, and comparing resulting performance and power consumption.

Best Poster Finalist (BP): no

Poster: PDF
Poster summary: PDF

Back to Poster Archive Listing