Work Package 01

Performance Assessment

This work package defines and improves SHAREing’s performance assessment framework by selecting relevant codes and systems, analysing assessment results to identify wider sector trends, and shaping a comprehensive methodology that can be applied across UK compute centres. It evaluates existing assessment practices, commissions studies to address methodological gaps, collaborates with training work packages on needed skills, and reviews the status of UK testbeds to establish best practices, consolidate essential information, and recommend future testbed developments that will best support progress in the community.

Work package lead Thomas Flynn

Eligibility: SHAREing tasks are open to all contributors based at UK universities or other organisations eligible for UKRI funding

WP1.1 – Performance Assessment

This working group identifies codes and systems that are accepted for SHAREing’s performance assessment exercises. The group ensures that SHAREing covers a wide variety of use cases and code and system characteristics. It also studies the delivered assessments to identify overarching trends and issues of the sector.

Community GPU codes - Task 036

See Details & Apply

Community CPU codes - Task 050

See Details & Apply

Reusable software for sustainable computing - Task 051

See Details & Apply

Suggest a new task based on WP1.1

Suggest

Example Preparation and Benchmarking of Community Codes using CASTEP

View

Investigation of quantum acceleration of the SWIFT HPC cosmology code

View

Applying the SHAREing performance assessment approach to community lattice quantum field theory software

View

FLAME GPU Benchmarking and Profiling for AMD

View

Testbed descriptions and consolidation

Results

Specification of Assessment Code Requirements

Results

Celeritas assessment

Results

Showcasing the Performance of Community GPU Codes in Life Sciences and Genomics

Results

WP1.2 – Assessment Methodology

This group studies and gathers performance assessment methodologies. The collected expertise helps SHAREing’s assessments, but the group also works towards a unique SHAREing approach which can be rolled out at multiple compute centres in the UK. For this, it has to assess shortcomings of existing methodologies, i.e. commission user studies, and work closely together with the training work packages whenever missing assessment skills are identified.

Machine settings - Task 045

See Details & Apply

Agent-based performance assessment - Task 046

See Details & Apply

Performance Modelling for Offloading - Task 048

See Details & Apply

Parallel Benchmark Games: A Community Challenge for Portable Research Software - Task 063

See Details & Apply

Performance analysis tools beyond C, C++ and Fortran - Task 064

See Details & Apply

Suggest a new task based on WP1.2

Suggest

Document DiRAC Performance Analysis Methodology

View

Python–Fortran Interoperability for Scientific Computing

View

Benchmark guidelines (kickstart)

Results

Supercomputing Wales

Results

WP1.3 – Testbeds

This group discusses the status quo of testbeds in the UK. We identify which information is to be collected and has to be consolidated, best practices in describing and assessing the testbeds, and the group also can make recommendations which testbeds they would like to see within the community to allow for faster progress in the field.

Using and optimising the Tenstorrent RISC-V system - Task 034

See Details & Apply

Deployment of LLMs on local infrastructure - Task 044

See Details & Apply

Optimising an HPC developer's environment - Task 047

See Details & Apply

Using and optimising the Monaka testbed - Task 049

See Details & Apply

Suggest a new task based on WP1.3

Suggest

Performance analysis for UK Numerical Relativity community codes

View

Live dynamic power management for HPC

View

HPC testbed and production system information service

Results

Documenting the HPC Hardware Lab at Durham University

Results