The International Conference for High Performance Computing, Networking, Storage, and Analysis
The International Conference for High Performance Computing, Networking, Storage, and Analysis
Authors: Maksim Melnichenko (University of Tennessee Knoxville, Innovative Computing Laboratory), Riley Murray (Sandia National Laboratories), Vivak Patel (University of Wisconsin Madison Department of Statistics), Weslley Da Silva Pereira (National Renewable Energy Laboratory (NREL)), Sherry Li (Lawrence Berkeley National Laboratory (LBNL)), Petros Drineas (Purdue University), Vasileios Georgiou (Purdue University)
Abstract: In recent years, randomized numerical linear algebra (RandNLA) proved to be more than a theoretical novelty: projects like RandLAPACK demonstrate its practical value across architectures, and projects like RandBLAS build trust in randomization as a tool for high-performance NLA. This BoF considers two main questions. First, what are the pressing issues in software standards and implementation that need to be resolved for RandNLA to become a core component of HPC? Second, how can we mobilize a community effort to make progress on these issues? The BoF will engage the audience to discuss the idea of growing the role of RandNLA in high performance computing and what it would take to scale from niche prototypes to robust, production-quality software libraries.
Long Description: Innovation in high-performance computing (HPC) often comes from unexpected places. While hardware trends have focused on ever-lower precision and architectural tuning to delay the impact of Moore’s law breakdown, another path forward has emerged from within theoretical computer science: Randomized Numerical Linear Algebra (RandNLA).
Initially viewed as a theoretical curiosity, RandNLA has matured significantly over the past two decades and has taken meaningful root in the HPC world through initiatives like BALLISTIC. The last few years have shown that RandNLA is not only practically viable but also capable of delivering substantial improvements across domains. Much of this progress has come from small-scale efforts: writing prototype software, stress-testing algorithms across domains, and building trust in randomization as a tool for performance and reliability, not uncertainty. We’re now at an inflection point: what began as small-scale innovation is ready to scale into a new generation of mainstream HPC software. But scaling up requires broader collaboration. Through open discussion, we hope to identify common needs across projects and platforms: standardized interfaces, GPU portability, dynamic memory models (e.g., via Kokkos), prospective integration pathways with frameworks like SLATE and Ginkgo, and sustainable development practices including release processes and test infrastructure.
This BoF aims to gather industry representatives, algorithm designers, software engineers, and HPC system architects to address these challenges collaboratively. We’ll reflect on what has been learned from recent efforts like the BALLISTIC initiative, which helped launch reusable software like RandBLAS (for sketching and sparse kernels) and RandLAPACK (for showcasing RandNLA implementations). These libraries have demonstrated that randomized algorithms can be both robust and performant, often outperforming traditional LAPACK routines by orders of magnitude.
Can we (and should we) treat RandNLA as foundational infrastructure in HPC, rather than as an add-on to classical algorithms and existing libraries?
What does an architecture-agnostic, high-performance RandNLA ecosystem look like?
How can we bridge the language gap, offering power users the C++ depth they need while maintaining accessible front-ends for Python and Julia users?
Can we consolidate the existing scattered efforts to build RandNLA software, and how can we reduce the fragmentation of the emerging efforts?
RandNLA use cases in production scientific codes
Lessons from designing object-oriented, extensible NLA libraries (e.g., RandLAPACK)
Future of standardized RandNLA kernels in vendor libraries
The BoF will begin with brief lightning talks from library developers, followed by an open-floor discussion. The goal is to form connections, share pain points and successes, and catalyze a longer-term working group or community roadmap.
Whether you're building software, developing algorithms, or looking to use RandNLA in your existing workflows, this BoF is a chance to help shape the future of randomized numerical computing in HPC.