Spanish Museum Builds Cray-1 Replica From 30 Mac Minis

The Museo de Historia de la Computación in Spain has unveiled a full-scale visual replica of the Cray-1 supercomputer, running on a cluster of 30 Mac Minis manufactured between 2012 and 2014, according to Ars Technica. The project, developed over 18 months, marks a 50th-anniversary tribute to the original Cray-1, which shipped in 1976, and to Apple's 1976 founding.

What hardware powers the replica?

The cluster runs on 30 Mac Minis that Ars Technica describes as "obsolete": five quad-core Intel Core i7 units and 25 dual-core Core i5 units. All but two of the machines date to 2012, with the remaining pair from 2014, per the outlet's reporting. Networked together, the museum's team orchestrates 64 active processing cores. The museum previously recorded a peak of 1.5 teraflops using all 70 available cores in the array before one node went offline, Ars Technica reported; the most recent tests the outlet cited put sustained performance at 1.3 teraflops across the 64 active cores.

Benchmarks were generated with a custom matrix-multiplication script written in Python, according to the report. That detail matters for interpreting the teraflops figures: they reflect performance on a specific, museum-authored workload rather than an industry-standard benchmark such as LINPACK, so the numbers describe this cluster's internal test rather than a certified supercomputer ranking.

How does this compare to the original Cray-1?

The 1976 Cray-1 had a peak theoretical performance of approximately 160 megaflops, Ars Technica reported, built around vector processing — a design that applies a single instruction to multiple data elements simultaneously. The Mac Mini replica instead uses a parallel cluster architecture, distributing work across independent processors that communicate over a network. The result, per the outlet, is consumer hardware now capable of far higher throughput than the purpose-built research machine it visually recreates. The museum uses the gap between the two architectures as a teaching tool, letting visitors watch matrix algorithms scale in real time from a single core up to the full 64-core array, according to the report.

What software connects the Mac Minis?

Rather than a standard Linux distribution, the project runs on macOS Mojave (10.14), which Ars Technica reported the team chose because it supports MPI (Message Passing Interface) natively. MPI is the software layer that lets each node in the cluster hand off and receive computing tasks from the others, which is the mechanism that turns 30 separate desktop computers into a single coordinated system rather than 30 machines running in isolation.

How did the museum cool and quiet the cluster?

Fan-cooled server racks are a known noise problem in data centers, and Ars Technica's report notes the builders avoided that outcome by mounting the Mac Minis in a steel rack chassis on aluminum trays and relying on passive convection cooling rather than fans.

"Compared to the roaring thunder of a modern data center, this cluster is nearly inaudible," Ars Technica reported.

The approach lets museum visitors stand next to a 64-core, multi-teraflop cluster without the noise typically associated with server infrastructure of that scale, according to the outlet.

How was the project assembled, and over what period?

According to Ars Technica's account, the build unfolded over roughly 18 months:

• Design phase: The team settled on a 1:1 scale visual reproduction of the Cray-1's exterior, including its signature padded bench seating, while sourcing used 2012–2014 Mac Minis instead of new hardware.
• Cluster assembly: Up to 70 Mac Minis were mounted and networked, with macOS Mojave and MPI chosen as the coordination layer between nodes.
• Cooling integration: Steel rack chassis and aluminum tray mounts were built to allow passive convection cooling in place of fan-driven systems.
• Testing: The team ran custom Python matrix-multiplication benchmarks, recording a peak of 1.5 teraflops on 70 cores before one node failed, and 1.3 teraflops on the 64 cores now active, per the outlet.
• Public launch: The finished replica went on display at the Museo de Historia de la Computación timed to the Cray-1's and Apple's 50th anniversaries in 2026.

Ars Technica's report frames the exhibit as a way for visitors to trace the shift from vector supercomputing to distributed, parallel computing using hardware many owners had already written off as outdated. The museum did not publish a separate technical paper alongside the exhibit, according to the available reporting, and the performance figures above come solely from the museum's internal benchmark script as described by Ars Technica.