Studying the Past to Predict the Future: The History of Computer Architecture Is Here

Computer Sciences professor Matt Sinclair

Computer Sciences professor Matt Sinclair, Computer Sciences graduate student Rutwik Jain, and their colleagues have analyzed 50 years of computer architecture – and are looking ahead to 50 more – in a recent International Symposium on Computer Architecture (ISCA) paper analysis. They call this work a “data-driven retrospective” and have drawn their data from the ISCA, which has been operating since 1973 (minus 1974, so annually since 1975) and is one of the longest running and most prestigious conferences in computer architecture. Moreover, Wisconsin has had a significant impact at ISCA, driving innovations that have influenced both the community and society at large. Five of the top 10 most prolific ISCA authors of all time are either Wisconsin faculty (current or emeritus) or obtained their PhD at Wisconsin. And UW-Madison has been a pioneer in Computer Architecture dating back to its very beginning, with Gene Amdahl completing his PhD on “The Logical Design of an Intermediate Speed Digital Computer” here in 1952.

Computer Sciences graduate student Rutwik Jain

Sinclair and his colleagues set out “to analyze the past 50 years of ISCA to understand who and what has been driving and innovating computing systems thus far.” They continue, “Our analysis identifies several interesting trends that reflect how ISCA, and Computer Architecture in general, has grown and evolved in the past 50 years, including minicomputers, general-purpose uniprocessor CPUs, multiprocessor and multi-core CPUs, general-purpose GPUs, and accelerators.”

In their analysis, Sinclair and the other authors note the remarkable growth in the annual number of papers submitted to ISCA – from about 70 in 1973 to over 400 in 2023. The number of authors per paper has also grown, highlighting the “increasing difficulty and challenges in building and researching increasingly large, more complex systems and the accompanying methodology required to perform the research.” The researchers also analyzed key words in abstracts, which papers were cited the most by decade, and the life cycle of its most-cited papers. 

How the number of submissions, accepted papers, and acceptance rate varied for the first 50 years of ISCA.

Paper topics evolved significantly through the decades. In the first 10 years of the ISCA conference, paper topics included network interconnects, fault diagnosis, and relational database systems – some for nascent multiprocessor systems while others focused on (at the time) emerging uniprocessors. The most cited paper was UW Electrical and Computer Engineering emeritus faculty member James Smith’s study on branch prediction strategies.

Heatmap of how topics appeared over the past 50 years for the Top 50 cited ISCA papers.

For the next decade (1983-1992), VLIW multiprocessors and multiple issue uniprocessors increased in popularity, and parallelism, cache coherence, and consistency rising in importance. Memory and cache architecture also emerged as topics. In the following decade, Sinclair and colleagues note the “boom and, to some extent, the bust in excitement around multi-core architecture.” And from 2013-22, dark silicon (work done by UW Computer Sciences professor Karu Sankaralingam, his student Emily Blem, and their collaborators), which started to emerge at the end of the previous decade, is “the natural next big research topic . . . specialization.” Sinclair notes, “ISCA’s fifth decade started with a top-cited paper on GPU power modeling and reconfigurable architecture, but subsequently Machine Learning dominates the rest of the list.”