ILLIAC IV pursued speed by spreading data across many processing elements. The Cray-1 reached a similar destination, doing the same operation across a lot of data, by a different road. Instead of many processors, it used one very fast processor with instructions that act on a whole vector of numbers in a single step. That choice made vector processing a commercial success and shaped supercomputing for years.

The idea

Vector processing: a single instruction operates on a whole array of numbers at once, keeping the processor busy instead of stalling between one scalar step and the next.

Vector registers

The Cray-1 carried eight vector registers, each holding 64 values. An ordinary scalar instruction operates on one number at a time, and between operations the processor often waits on memory. A vector instruction instead applies one operation across all 64 values in a register, feeding them through the arithmetic units in a steady stream so the processor stays busy rather than stalling. For work that repeats the same calculation over long arrays, this is a large and direct gain.

Why it won

The Cray-1 was designed by Seymour Cray and was the first supercomputer to make the vector design a practical, commercial success. The first system was installed at Los Alamos National Laboratory in 1976. Its speed came from the combination of vector processing and a tightly engineered design built for raw clock speed.

Legacy

The vector idea did not fade. It reappeared as the SIMD instructions in mainstream CPUs and as the core execution model of graphics processors, where the same operation runs across many data elements at once. The Cray-1 is where that approach first proved it could win.

Sources

  • “Cray-1,” Wikipedia. https://en.wikipedia.org/wiki/Cray-1 . Supports Seymour Cray as architect, the Cray-1 as the first supercomputer to successfully implement the vector processor design, the eight 64-element vector registers, and the first system installed at Los Alamos National Laboratory in 1976.