Periodic Reporting for period 1 - POP3 (Performance Optimisation and Productivity 3)
Periodo di rendicontazione: 2024-01-01 al 2025-06-30
The Centre of Excellence on Performance and Productivity Optimization (POP CoE) has the mission of assisting a wide community of HPC application developers and users across scientific and industrial domains. Its goal is to help them identify and understand performance issues in their applications, thereby improving efficiency and productivity. This is achieved through independent performance audits of codes, providing both qualitative and quantitative insights using POP’s tools and methodology.
While POP3 continues to target applications at all scales, special emphasis is placed on large-scale executions, with the objective of improving the scalability and efficiency of European flagship codes identified by other Centres of Excellence (CoEs) across diverse scientific areas. To achieve this, POP3 advances its performance tools to ensure broad usability across EuroHPC systems and to enable analysis at the scale required by these applications.
The main objectives of POP3 are twofold:
- Advancing tools and methodology:
POP3 develops performance analysis tools and techniques that address key aspects of scalability, modeling, and methodology. These are adapted to fully support EuroHPC platforms, ensuring they can handle large-scale analysis. The tools remain open source and are complemented by a comprehensive training program.
- Assessing applications at scale:
POP3 provides performance assessments for users and developers running, or aiming to run, large-scale applications. These assessments include recommendations on best practices for parallel programming and propose refactoring strategies to improve efficiency—covering both hardware utilization and energy consumption.
The expected impact of POP3 is threefold:
- Advancing the transition towards exascale:
By assessing a large number of codes, POP3 helps developers identify scalability bottlenecks and provides strategies to overcome them. The transition to exascale is both a technical and cultural challenge, requiring not only high-profile success stories at the largest scales but also continuous improvements across all scales. A solid understanding of execution behavior is essential to steer applications effectively toward higher scalability.
- Enabling effective applications:
Efficiency lies at the core of the POP3 methodology. By measuring execution and scaling efficiencies, POP3 offers clear guidance on maximizing resource utilization while minimizing time-to-solution and energy consumption. Although the project does not address the scientific quality of applications—that remains the role of domain-specific CoEs or industrial stakeholders—POP services ensure codes deliver results more effectively, improving the ratio of scientific or industrial output per unit of resources.
- Addressing scientific, industrial, and societal challenges:
POP3’s impact spans research, academia, industry, and public administration. Scientific research often pushes the boundaries of scalability, while industry typically operates at smaller scales. POP3 supports both communities at their respective levels, preparing them for the next scale. By disseminating a culture of efficiency and best practices, POP3 creates value not only for individual users but also for society, through more sustainable and cost-effective use of HPC resources.
While POP3 continues to target applications at all scales, special emphasis is placed on large-scale executions, with the objective of improving the scalability and efficiency of European flagship codes identified by other Centres of Excellence (CoEs) across diverse scientific areas. To achieve this, POP3 advances its performance tools to ensure broad usability across EuroHPC systems and to enable analysis at the scale required by these applications.
The main objectives of POP3 are twofold:
- Advancing tools and methodology:
POP3 develops performance analysis tools and techniques that address key aspects of scalability, modeling, and methodology. These are adapted to fully support EuroHPC platforms, ensuring they can handle large-scale analysis. The tools remain open source and are complemented by a comprehensive training program.
- Assessing applications at scale:
POP3 provides performance assessments for users and developers running, or aiming to run, large-scale applications. These assessments include recommendations on best practices for parallel programming and propose refactoring strategies to improve efficiency—covering both hardware utilization and energy consumption.
The expected impact of POP3 is threefold:
- Advancing the transition towards exascale:
By assessing a large number of codes, POP3 helps developers identify scalability bottlenecks and provides strategies to overcome them. The transition to exascale is both a technical and cultural challenge, requiring not only high-profile success stories at the largest scales but also continuous improvements across all scales. A solid understanding of execution behavior is essential to steer applications effectively toward higher scalability.
- Enabling effective applications:
Efficiency lies at the core of the POP3 methodology. By measuring execution and scaling efficiencies, POP3 offers clear guidance on maximizing resource utilization while minimizing time-to-solution and energy consumption. Although the project does not address the scientific quality of applications—that remains the role of domain-specific CoEs or industrial stakeholders—POP services ensure codes deliver results more effectively, improving the ratio of scientific or industrial output per unit of resources.
- Addressing scientific, industrial, and societal challenges:
POP3’s impact spans research, academia, industry, and public administration. Scientific research often pushes the boundaries of scalability, while industry typically operates at smaller scales. POP3 supports both communities at their respective levels, preparing them for the next scale. By disseminating a culture of efficiency and best practices, POP3 creates value not only for individual users but also for society, through more sustainable and cost-effective use of HPC resources.
During this reporting period, the flagship tools of the POP3 project have been significantly extended to enhance compatibility with EuroHPC platforms. Particular emphasis was placed on supporting the latest AMD GPUs, while also strengthening existing support for NVIDIA GPU architectures.
In parallel, the complete suite of POP3 performance analysis tools has been progressively deployed as system modules across EuroHPC infrastructures. This effort ensures broad accessibility for all EuroHPC users. To date, most tools have been successfully deployed on five to six EuroHPC platforms and are now publicly available.
On the service side, 31 codes were assessed during the reporting period, complemented by 7 second-level services. Of these, 30 activities (25 assessments and 5 second-level services) were delivered to customers affiliated with ongoing Centres of Excellence (CoEs). All assessments provided developers with actionable insights on how to refactor their codes to improve scalability and efficiency. Strong collaboration with the tools development task has ensured that the tools evolve in step with the platforms and code features being analyzed.
The project also defined the Customer Advocacy methodology and implemented new questionnaires in the EUSurvey platform for: Performance Assessments (PA), Second-Level Services (including PoC, energy efficiency, correctness checks, and advisory studies), and Return on Investment (ROI).
To date, 79% of PA surveys have been completed, revealing high user satisfaction: 83% of respondents reported being “Very satisfied” or “Satisfied.” Insights from these surveys have been shared with relevant experts and uploaded to the project Wiki.
In terms of capacity building, 7 training events on tools and performance analysis methodology were delivered.
Finally, the co-design materials hosted on the dedicated webpage have been continuously updated. Additions include 4 new kernels and 7 new technical pages, alongside improvements in content structure and navigation to enhance user experience.
In parallel, the complete suite of POP3 performance analysis tools has been progressively deployed as system modules across EuroHPC infrastructures. This effort ensures broad accessibility for all EuroHPC users. To date, most tools have been successfully deployed on five to six EuroHPC platforms and are now publicly available.
On the service side, 31 codes were assessed during the reporting period, complemented by 7 second-level services. Of these, 30 activities (25 assessments and 5 second-level services) were delivered to customers affiliated with ongoing Centres of Excellence (CoEs). All assessments provided developers with actionable insights on how to refactor their codes to improve scalability and efficiency. Strong collaboration with the tools development task has ensured that the tools evolve in step with the platforms and code features being analyzed.
The project also defined the Customer Advocacy methodology and implemented new questionnaires in the EUSurvey platform for: Performance Assessments (PA), Second-Level Services (including PoC, energy efficiency, correctness checks, and advisory studies), and Return on Investment (ROI).
To date, 79% of PA surveys have been completed, revealing high user satisfaction: 83% of respondents reported being “Very satisfied” or “Satisfied.” Insights from these surveys have been shared with relevant experts and uploaded to the project Wiki.
In terms of capacity building, 7 training events on tools and performance analysis methodology were delivered.
Finally, the co-design materials hosted on the dedicated webpage have been continuously updated. Additions include 4 new kernels and 7 new technical pages, alongside improvements in content structure and navigation to enhance user experience.
During the first reporting period, POP3 has delivered significant results with direct scientific, economic, and social impact. Scientifically, 31 codes were assessed, 25 of them from other Centres of Excellence, providing developers with actionable insights to improve scalability, portability, and sustainability, thereby enabling more accurate and reliable simulations. POP tools have been extended to support diverse EuroHPC platforms, with enhanced compatibility for AMD and NVIDIA GPUs, while the POP methodology is being adapted for heterogeneous accelerator-based systems. Co-design kernels developed within the project are actively supporting collaborations with European processor initiatives such as EPI, EUPilot, DARE, and EUPEX. Seven peer-reviewed publications and multiple training events further strengthen dissemination and knowledge transfer.
Economically, the open-source tools released by POP3 reduce reliance on costly, vendor-specific solutions and empower a broad HPC user base. Performance assessments lead to indirect impact by reducing execution time and computational costs when optimizations are adopted.
Socially, POP3 contributes to energy efficiency by promoting better resource utilization, shorter runtimes, and reduced CO2 emissions. Two dedicated energy-efficiency studies demonstrate tangible benefits, particularly for organizations running large-scale workloads. Together, these results foster scientific excellence, industrial competitiveness, and sustainable HPC usage in Europe.
Economically, the open-source tools released by POP3 reduce reliance on costly, vendor-specific solutions and empower a broad HPC user base. Performance assessments lead to indirect impact by reducing execution time and computational costs when optimizations are adopted.
Socially, POP3 contributes to energy efficiency by promoting better resource utilization, shorter runtimes, and reduced CO2 emissions. Two dedicated energy-efficiency studies demonstrate tangible benefits, particularly for organizations running large-scale workloads. Together, these results foster scientific excellence, industrial competitiveness, and sustainable HPC usage in Europe.