SC26 Committee Honored GPU Computing Paper
The seminal 2009 work on sparse matrix-vector multiplication has been awarded the SC26 Test of Time Award.
Updated on Sept. 29, 2026 in Mathematics
Live Poll
Should scientific research be evaluated primarily by its long-term influence rather than immediate project results?
The SC26 committee has selected the 2009 research paper "Implementing Sparse Matrix-Vector Multiplication on Throughput-Oriented Processors" for its prestigious Test of Time Award. This award recognizes scholarly contributions that continue to influence high-performance computing long after their initial publication.
Why it matters
The research, co-authored by Nathan Bell and Michael Garland, established foundational techniques for accelerating irregular data computations on GPUs. Its ongoing relevance reflects the central role of sparse matrix operations in modern parallel computing and machine learning workloads.
The study utilized an NVIDIA GeForce GTX 285 graphics card to evaluate sparse matrix formats including DIA, ELL, CSR, and COO. It demonstrated methods to organize computation to minimize execution and memory divergence.
The players
Nathan Bell
A Principal Engineer at Google who co-authored the seminal 2009 research paper.
Michael Garland
The Senior Director of Programming Research at NVIDIA who focuses on GPU computing architectures.
The details
The researchers addressed the challenges of performing sparse matrix-vector multiplication—a mathematical operation where a matrix contains mostly zeros—on hardware designed for dense data. By optimizing how data is mapped to threads, they mitigated the performance penalties associated with irregular memory access patterns common in large, sparse datasets. This work allowed developers to better utilize throughput-oriented processors, which are architectures designed to execute many small operations in parallel.
Timeline
2006: Michael Garland joined NVIDIA.
2009: The original research paper was presented at SC09 in Portland, OR.
September 29, 2026: The Test of Time Award was officially announced.
The Tech Race
This award honors a milestone in the long-term competition to optimize GPU performance for diverse mathematical applications. The recognition follows a consistent trajectory within the SC community to elevate work that successfully transitioned from niche academic research to standard industry practice.
This research provides the fundamental logic used in modern GPU libraries for handling sparse data structures. Software engineers working in machine learning and scientific simulation rely on these optimized matrix formats to achieve efficient hardware utilization today.
The takeaway
The sustained impact of this 2009 paper underscores the importance of algorithmic efficiency in parallel processing. Readers interested in the future of these techniques should monitor the upcoming session at SC26 for updates on GPU sparsity trends.
What happens next
Michael Garland is scheduled to present a talk titled "Twenty Years of Sparsity on GPUs" at the upcoming SC26 conference.
Further reading
Explore the evolution of computational methods in our Mathematics archive.
More information
Read the Original research paper access to understand the original benchmark methodology.
Source note: This article includes information reported by HPCwire.
Live Poll
Should scientific research be evaluated primarily by its long-term influence rather than immediate project results?
