30 "Obsolete" Mac Minis Recreate Iconic Cray-1 Supercomputer Replica

The Museo de Historia de la Computación in Spain has unveiled an impressive 1:1 scale visual replica of the groundbreaking Cray-1 supercomputer, originally released in 1976. This ambitious project, developed over an 18-month period, serves as a dual commemoration: celebrating the 50th anniversary of the Cray-1's introduction and honoring the founding of Apple. Instead of employing costly, state-of-the-art technology, the creative team behind the replica ingeniously powered the system using 30 Mac Minis dating back to 2012. While these machines might be considered "obsolete" by today's standards, when networked together, they form a powerful cluster. In recent tests, the developers successfully orchestrated the 64 active processing cores within these Mac Minis to achieve a combined performance of 1.3 teraflops. Previously, the museum reported reaching 1.5 teraflops by utilizing all 70 available cores before a single node experienced an unexpected outage. These performance benchmarks were meticulously obtained through the execution of a custom matrix-multiplication Python script. The cluster is comprised of five PCs equipped with quad-core Intel i7 processors, while the remaining 25 units feature dual-core Intel i5 processors. The vast majority of these Mac Minis originate from 2012, with only two units being slightly newer, from 2014. This meticulously crafted replica offers museum visitors an invaluable and tangible educational tool, enabling them to grasp the profound shifts in computer architecture that have occurred since 1976. The original Cray-1, a pioneering machine developed by Seymour Cray's Cray Research, was renowned for its innovative use of vector processing. This technique allowed a single instruction to be applied simultaneously to multiple data elements, significantly accelerating scientific and engineering computations. In stark contrast, the replica operates on a parallel cluster model, distributing computational tasks across its numerous independent processors. The peak theoretical performance of the original Cray-1 supercomputer was approximately 160 megaflops, a figure that is dwarfed by the capabilities of the re-created system. This project powerfully illustrates how readily available consumer hardware, when intelligently networked and configured, can emulate and even surpass the performance of highly specialized, high-performance machines from earlier eras. It represents a remarkable evolution from purpose-built research systems to the distributed power of modern consumer technology. The fundamental difference between the Cray-1's vector processing paradigm and the replica's parallel cluster architecture vividly underscores the significant changes in how computational power is harnessed and managed, offering a practical and accessible demonstration of parallel computing principles to a broad public audience.