Key Computer Labs was a supercomputer start-up in Silicon Valley in the late 1980s. The founders (Jeff Rubin and Tom McWilliams) were veterans of the Livermore Labs' S1 project, who had gone on to make a bit of money in commercializing the design tools (SCALD, if I remember correctly) that had originally been developed for the S1. Key raised enough capital to get going, but soon found themselves short enough of cash that they were rescued by Amdahl. It has been speculated that this was an insurance policy against IBM delivering new technology into the mainframe marketplace sooner that Amdahl expected. Amdahl ultimately stopped the supercomputer project and turned Key into something of an advanced architecture lab.
Key's one moment of glory came in the late 1980s, after the Amdahl acquisition, when their CAD mainframes found a new prime number. The fact that they had that many cycles to spare spoke volumes about the state of the project.
The limited information presented here is based on conversations during job interviews at Key both before and after the Amdahl acquisition. As in all of these pages, I would prefer to have first-hand information from project veterans. Please send any information to KevinK@acm.org.
The Key machine was to have been a heavily pipelined design, with a form of predicated execution as a mechanism to avoid branch bubbles.
The Key machine was to have been built in semi-custom ECL, packaged in dense thermal-exchange modules.
The Key machine was originally to have run a port of UNIX System V.
They seemed to be focusing on cost-reducing the least expensive component of a supercomputer: the CPU.
Supercomputing is extremely capital-intensive, and the time to break-even as a business is very long relative to the time-scales to which venture capitalists have grown accustomed.
I'd love to see one, but I don't know if any ever existed.