From Paper to Powerhouse: Three Big Ideas That Stand the Test of Time

Technology is constantly pushing forward to deliver innovations that improve lives and create optimal outcomes. But in the pursuit of what’s next, it can be easy to overlook the accomplishments that serve as the foundation for innovations that we enjoy today. That’s precisely why, in 2013, the SC Conference established the Test of Time Award, which highlights the groundbreaking research whose value only deepens with the passing years. 

In 2013, as General Conference Chair for SC13, Bill Gropp helped usher the Test of Time Award into existence. “Franck Cappello, a member of the Technical Program Committee, proposed the idea,” he told us. “We all thought it was a great way to recognize the lasting impact of research from the SC community.” Today, Bill Gropp continues to contribute to the award, serving as the SC25 Test of Time Award Chair. He is responsible for leading the selection process and championing the recognition of the research that continues to impact the HPC/AI field.

The SC25 Communications team spoke with Bill to explore his thoughts on three landmark achievements he considers among the most influential in shaping today’s computing landscape. Here are his personal selections.

Divide and Conquer

In Plain English

This paper showed how dividing big data problems into smaller parts helped supercomputers work faster and more efficiently, unlocking a new level of scale for tackling the world’s most complex challenges. It has now become a foundational technique for modern high performance computing.

Summary of the Advancement

At SC14, Bruce Hendrickson and Rob Leland were honored for their influential 1995 paper, “A Multi-Level Algorithm for Partitioning Graphs.” Their work transformed how large problems are divided and distributed across processors, breaking them into more manageable sections that can run simultaneously across multiple processors. Before their innovative paper, distributing workloads effectively across processors was incredibly challenging, often limiting the scale and speed of computations. However, the duo’s multi-level algorithm streamlined the partitioning process, dramatically enhancing performance and setting the stage for more ambitious computing tasks.

Why it Matters Today

Today, Hendrickson and Leland’s innovation quietly powers innovations in virtually every field that supercomputing touches, including engineering, climate science, aerospace, pharmaceuticals and artificial intelligence. Without efficient partitioning, complex computing tasks — such as modeling the climate or designing safer aircraft — would take significantly more time and resources.

Bill Gropp Says

“Before this paper, setting up a complex problem could actually take more time than solving it. Hendrickson and Leland’s algorithm didn’t just solve that — they provided a new way of thinking that is still fundamental today. Practically every field relying on large-scale computing still benefits from their approach.”

Code That Optimizes Itself

In Plain English

Dongarra and Whaley taught code to automatically fine-tune itself, adjusting its own performance to run at maximum speed on any machine without manual adjustments.

Summary of the Advancement

At SC16, Jack Dongarra and Clint Whaley received the Test of Time Award for their 1998 paper, which introduced ATLAS — an innovation that addressed a growing challenge in computing: hardware had become so complex that manually tuning software for optimal performance was becoming nearly impossible.

At the time, getting software to run efficiently on emerging hardware required weeks or even months of expert tuning. The ATLAS project changed that by automating the optimization process. ATLAS systematically generated variations of performance-critical code, tested them, and selected the best-performing version for a given system. The result was software that was faster, cheaper to run, and immediately adaptable to new hardware technologies as they emerged. 

Why it Matters Today

Automatic tuning is now a foundational concept in high-performance software, driving efficiency across every field. Today’s AI models, cloud computing systems, pharmaceutical research platforms, and climate simulations all depend on the self-optimizing approach pioneered by ATLAS.

Bill Gropp Says

“Before ATLAS, getting high performance required weeks, or even months, of manual tuning by expert programmers. Dongarra and Whaley automated that, creating a smarter, faster, self-optimizing approach that continues to power nearly every field touched by high performance computing today.”

Beyond Graphics

In Plain English

This paper showed, before it was cool, that graphics processors (GPUs) could be used for serious scientific computing, not just video games. Their insights helped spark a revolution that paved the way for today’s increasingly AI-powered world.

Summary of the Advancement

At SC24, Zhe Fan, Feng Qiu, Arie Kaufman, and Suzanne Yoakum-Stover were recognized with the Test of Time Award for their game-changing 2004 paper, “GPU Cluster for High Performance Computing.” While the power of GPUs is practically taken for granted today, at the time, they were viewed as niche hardware designed for rendering graphics, rather than as serious computational tools. This quartet of researchers challenged that assumption, demonstrating that GPUs could be clustered and used to efficiently run scientific workloads. Their work dramatically shifted perceptions about the potential of GPUs. Even early-generation GPUs, they showed, could significantly accelerate simulations, making a compelling case for their broader use in HPC. 

What’s more, their insight was timely. The paper appeared just before the end of Dennard scaling — a shift that meant processors would no longer get faster without embracing parallelism. By proving that GPUs could tackle a wide range of complex problems, this paper helped the community rethink what was possible in high-performance architectures.

Why it Matters Today

GPU-accelerated computing is now a cornerstone of nearly every major breakthrough in AI, climate modeling, drug discovery, and engineering. This paper helped kick off a revolution in computing architecture — one that continues to shape our digital world more than two decades later.

Bill Gropp Says

“Early GPUs were very limited and difficult to program, but this paper clearly showed they could handle real, meaningful science. That demonstration was critical. It made people realize that GPUs need to be taken seriously, laying the foundation for today’s GPU-driven world.”

Nominate a Paper Today