NAISS
SUPR
NAISS Projects
SUPR
Auto-tuning SYCL benchmarks and applications on AMD GPUs
Dnr:

NAISS 2025/22-1569

Type:

NAISS Small Compute

Principal Investigator:

Hari Abram

Affiliation:

Chalmers tekniska högskola

Start Date:

2025-11-14

End Date:

2026-12-01

Primary Classification:

20206: Computer Systems

Webpage:

Allocation

Abstract

This project aims to evaluate an auto-tuning tool designed to explore compile-time and runtime parameters for identifying optimal performance configurations on GPUs. Current experiments are conducted on NVIDIA GPUs; however, to achieve a comprehensive performance comparison, I plan to extend the evaluation to AMD GPUs. Access to AMD GPUs on the Dardel supercomputer is requested to conduct these experiments and analyze cross-vendor performance characteristics.