Currently, in addition to many general-purpose high-performance computing (HPC) and GPU resources available for the advanced research computing (ARC) community, there are novel accelerator hardware-based supercomputers and resources that are being built and made available to the science and engineering community. Recent deployments include machines with purpose-built processors from ARM, Habana Labs, Cerebras, Graphcore, BittWare, SambaNova, etc. These resources provide specialized architectures for low-latency, power-efficient, and AI models and operators of HPC and Big Data applications. They may also leverage Liqid or GigaIO’s unique composable (software defined hardware reaggregation) capabilities. In some cases, specific processor architectures are meant for optimal training versus inference, high bandwidth memory systems optimized for AI applications, and specific interconnects designed for scaling AI applications. The software stack on the resources typically features standard high-level frameworks such as PyTorch and TensorFlow with customized compilers, graph engines, and runtimes that can optimally utilize the specialized hardware. Scientific applications built using these custom frameworks allow researchers to transparently take advantage of the custom hardware. In these respects, these machines are different from general purpose CPU and GPU based machines where the software stack is generalized to support a large set of applications. Generally available standardized techniques and community frameworks at the high level, eases the transition of researchers to porting and using these new machines. In addition to offering tremendous performance gains by transparently taking advantage of specialized architecture, these generally applicable frameworks, coupled with purpose-built tools help researchers seamlessly transition their applications and scientific workflows between generalized HPC (CPU+GPU) environments and these specialized CI resources.

We will discuss the ideas behind such hardware, project plans that are developed jointly with vendors of these systems, and the architecture of such machines including processor cores, accelerators, interconnects, file systems, etc. We will give insights into how the systems are managed, what cluster management tools and approaches are used, what system monitoring tools are used, how the I/O subsystems are architected and managed, what batch systems are used, and how user accounts and allocations are managed for researchers. We will discuss what experiences from managing and running general-purpose HPC and GPU machines are carried over to manage and support these novel testbeds versus what is new regarding capabilities, system software, scheduling software, and policies. Similarly, from researchers and application support perspective, we will discuss how these machines are benchmarked, how users are onboarded, how researchers are applying workflows that are utilizing the machines, training and enabling researchers to use these machines effectively, and how the users have been able to port, run and optimize applications using frameworks such as PyTorch and TensorFlow. We will share researcher’s experience regarding using this specialized hardware versus regular general-purpose HPC and GPU machines. We will discuss how the staff of the centers providing such machines and their vendors collaborate to enable the user communities of such specialized and innovative architecture machines. Our experience is that there is close and ongoing collaboration needed among three parties – the centers’ system and applications staff, the experts from the vendors, and the user community – to onboard and effectively utilize these machines for the science and engineering community; this is a different approach than what is needed for general purpose HPC and GPU machines that have commodity processors and are in use within the supercomputing community for a long time. In addition to these, the allocation of such novel and evolving hardware may require adaptation of new allocation and accounting policies.

Authors: David Hart, Nathan Tolbert, Rob Light and Stephen Deems | The eXtensible Resource Allocation Service (XRAS), a comprehensive allocations environment for managing the submission, review, and awarding of resource allocations has been essential to managing the resource allocation needs of the national cyberinfrastructure for the past decade. As a software-as-a-service platform, XRAS supports not only the needs of its primary stakeholder program, but also the processes of half a dozen other national and regional resource providers. With flexible mechanisms that allow clients to integrate with the platform, configuration options to meet the needs of many process options, integration with the ORCID ecosystem, and a fee-for-service approach to sustainability, XRAS has continuously improved on its original feature set and added new features to better meet the needs of its clients. Today, XRAS supports a core workload of 500 allocation requests each quarter, and the development roadmap for the system is focused on expanding the types of resources XRAS supports, enabling resources to be integrated in novel ways, and increasing the number of resource federations that XRAS can support.

Join the Precision Convergence Collaboration (PCC) as we discuss our efforts to explore how advanced research computing and data can contribute to the scientific understanding of multi-scale mechanisms that connect human biology and brain function to human behavior and social and economic decision-making. This requires the development of a digital ecosystem bridging laboratory and real-world science, policy, and innovation. By leveraging cutting-edge computing resources and collaborative frameworks, the PCC aims to create a dynamic platform for interdisciplinary collaborations. Since 2021, the PCC has been evolving around a series of webinar panel sessions. The discussions seeded by these sessions suggest that the intersection of neuroscience, economics, healthcare, and technology is essential for the study of contextual decision-making that includes considerations for health, social, and environmental factors. The further development, implementation, and validation of these ideas require significant broadening of the community of researchers and stakeholders who would participate in the PCC. The goal of this panel is to share our process and ideas with the advanced computing community, particularly with under-represented and under-resourced institutions, attracting new partners to build PCC into a sustainable community of practice.

This panel will discuss a number of dimensions of energy efficiency and its importance in research computing. The panel discussion will be based around a forthcoming position paper on energy efficiency from a working group of the Coalition for Academic Scientific Computation (CASC). The goals of the position paper are to highlight the main avenues to energy savings, namely, through hardware and software. Energy-aware hardware has a direct impact on efficiency. Hardware vendors need to invest in exploring more energy-efficient ways of execution and make use of advanced technology that is not mainstream. On the software side, software can have a large impact on energy use, but that requires software to be more “energy aware” and tools and frameworks must exist to easily, and in a standardized way, provide energy use as a new performance measure in addition to traditional performance metrics.

Authors: Dana O’Connor and Paola A. Buitrago | Predicting the properties of molecular crystals is imperative to the field of materials design. In lieu of alternative methods, advances in machine learning have made it possible to predict the properties of materials before synthesis. This is especially important for organic semiconductors (OSCs) that are prone to exhibit polymorphism, as this phenomenom can impact the properties of a system, including the bandgap in OSCs. While graph neural networks (GNNs) have shown promise in predicting the bandgap in OSCs, few studies have considered the impact of polymorphism on their performance. Using the MatDeepLearn framework, we examine five different graph convolution layers of ALIGNN, GATGNN, CGCNN, MEGNet, and SchNet, which all have graph convolutions implemented in torch geometric. A dataset of functional organic molecular crystals is extracted from the OCELOT database, which has calculated density functional theory (DFT) values for the bandgap as well as several sets of polymorphs. The trained models are then evaluated on several test cases including the polymorphs of ROY. In future work we plan to examine the impact of graph representations on the performance of these models in the case of predicting properties of polymorphic OSCs.

Authors: Maureen Dougherty, Barr von Oehsen, Jason Kaelber, Jeremy Schafer, Bala Desinghu, Morgan Ludwig and John Goodhue | The Ecosystem for Research Networking (ERN) CryoEM Remote Instrument Access Pilot Project addresses the challenges and barriers facing the transition of instrument-driven science self-contained islands to federated wide-area internet accessible instruments. This project’s goal is to facilitate cross-institutional infrastructure sharing at the interface of computing and electron microscopy through a web portal leveraging federated access to one or more instruments, workflows/pipelines utilizing edge computing in conjunction with advanced computing, along with real-time monitoring for experimental parameter adjustments and decision making. The intent is to foster team science and scientific innovation, with emphasis on under-represented and under-resourced institutions, through the democratization of scientific instruments. This paper presents a short summary some of the challenges encountered with the latest development efforts of this active project, creating a FABRIC Cloudlet and incorporating Pittsburgh Supercomputing Center’s Bridges-2 as an advanced computing option.

Over the last several years, OpenHPC has emerged as a community-driven, reference stack providing a variety of common, pre-built software, libraries and services to deploy and manage an HPC Linux cluster. Formed initially in November 2015 and formalized as a Linux Foundation project in June 2016, OpenHPC has been adding new software components and now supports multiple OSes/architectures. At this BoF, speakers from the OpenHPC Technical Steering Committee will provide updates on the latest 2.x release introduced in the first quarter of 2024. We will then discuss the latest release of OpenHPC 3.x, which introduces support for RedHat 9.3, and SUSE Leap 15.6. Finally, we will invite open discussion giving attendees an opportunity to provide feedback on current conventions, packaging, request additional components and configurations, and discuss general future trends.

The Cyberinfrastructure Professional (CIP), as described by the National Science Foundation (NSF), refers to the “community of individuals who provide a broad spectrum of skills and expertise to the scientific and engineering research enterprise by inventing, developing, deploying and/or supporting research CI or CI users. Examples of CIPs include CI system administrators, CI research support staff, CI research software engineers, data curators, and CI facilitators, and may also include computation research scientists and engineers who are not traditional academic paths.”The CIP roles regularly elevated within our community include system administrators, research software engineers, computation research scientists, and facilitators, other vital roles need to be recognized. These roles are essential and critical to the drive, structure, and impact of our CI community and include expertise in program and project management; outreach and engagement; communications; training and education; facilitation; and business operations including finance and human resourcesIndividuals in these roles are adept problem solvers, skilled communicators, and expert planners, contributing a diverse array of crucial skills to organizational success. They excel in coordinating programs and attention to detail, ensuring precision and dependability. Building and nurturing relationships for collaboration comes naturally, alongside exceptional communication and listening abilities, facilitating clear exchanges of ideas and garnering support. Their strategic thinking enhances comprehension of the broader picture, enabling effective project management and alignment with overarching goals. Additionally, their empathetic and nurturing nature fosters positive work environments, while their ability to collaborate across disciplines and uphold institutional knowledge ensures continuity and informed decision-making.These vital CI professionals undertake a myriad of responsibilities and roles within their purview. They often sit on governance or advisory boards, chair committees, spearhead strategic planning efforts, and craft organizational goals. Additionally, they lead or contribute to sponsored projects, coordinate outreach endeavors to democratize access to CI resources, and oversee various communication channels, from internal team communications to public-facing marketing materials. Their role extends to conference management for events like PEARC and SC, organizing student programs, and managing organizational operations, including budgeting and staff oversight. They facilitate discussions and engagements, manage complex projects, develop educational content such as curriculum, workshops, and training materials, and handle software tools and event logistics, ensuring seamless execution across all fronts.Our objectives for this panel are to 1) identify the variety of important CIP roles that are critical to the success of the cyberinfrastructure community, 2) raise awareness of the vital contributions and impact of diverse CIP roles, and 3) discuss these diverse career opportunities and how our panelists arrived in their roles. This panel discussion is dedicated to fostering inclusion and a sense of belonging by highlighting the indispensable roles that contribute to the success of the CI field. Ultimately, we all have the same goal: to support the research objectives of our computing centers, higher education, and research institutions, while also furthering the mission of democratizing access to cyberinfrastructure resources.

In this Birds of a Feather ssession, the ACCESS Resource providers (RP) will give a brief overview of the available resources and their unique characteristics. The presentation portion of this BoF will highlight the variety of available resources and will be followed by a discussion with the community, allowing the audience to directly interact with the RP representatives. We hope to seed the discussion with topics but allow attendees to steer the discussion, perhaps uncovering topics not suggested here.

Coastal Louisiana is facing potentially devastating land loss, vegetation change, and storm surge-based flooding over the next fifty years. Our team at the Pittsburgh Supercomputing Center (PSC) is partnering with the Coastal Protection and Restoration Authority (CPRA) to develop web-based visualization software that supports their efforts to model mitigation projects and to develop a coastal master plan. Like the coastal environment, the software is evolving to handle expanding datasets and to provide more powerful and flexible visualizations.In this poster, we describe three software applications we built to support the 2023 Coastal Master Plan, highlighting the ways in which their architecture evolved to meet emerging needs from the research team and the plan’s other audiences. We also describe our ongoing work to discover and define requirements for the next generation of software for the 2029 Coastal Master Plan. We believe the partnership between CPRA and PSC exemplifies “human-powered computing,” where the needs of scientists, policymakers, and residents of coastal Louisiana drive the technical direction of the software.

Neocortex is an National Science Foundation (NSF)[7] system for artificial intelligence workflows that integrates hundreds of thousands of cores coupled with high-speed on-chip memory, making it ideal for complex AI tasks that require high throughput and rapid processing. It represents a significant advancement in computational infrastructure and is currently transitioning from an experimental platform to a full-scale production system. This abstract serves as a call for applications from researchers across disciplines looking to leverage this state-of-the-art resource for groundbreaking AI-related projects. The system is accessible through the ACCESS website, inviting all research projects and compatible models that could benefit from the unique capabilities of Neocortex.