It is a touch misleading to refer to a single company by the name of Cray, as there have been two companies named Cray that were run by Seymour Cray, and one unrelated networking company. In this node, I will not be covering the relatively unimportant networking company, but will instead be focusing on Cray Research Incorporated (CRI), and Cray Computer Corporation (CCC). Seymour Cray also was involved in the design of large machines at Control Data Corporation (CDC), but those machines are not classified as Cray’s. CRI is however, the traditional Cray, as CCC failed before their machines really reached the market.
Cray machines can easily be categorized into a number of separate classes, based upon architecture, time, and the developing company. The primary distinction is that between the vector based super computers, which are more traditional machines for the company, and the MPP machines that they have been building more recently. This distinction is more then just technological, it also has somewhat of an ideological separation, as Seymour disliked clusters and MPP based designs, at one point questioning, “If you were plowing a field, which would you rather use? Two strong oxen or 1024 chickens?” As a result, the MPP machines, such as the T3e, T3d, and CS6400 were designed following his departure from CRI. I am unclear as to whether or not he changed his position on supercomputer design late in his life, as he gave indication of doing so, but I can find no comments on it prior to his death.
The vector designs of Cray can further be divided into categories; this separation is mostly one of marketing, and packaging, rather then technology. Prior to the mid-late ‘80s, Cray had a unified line of machines, however, around this period a small company began building supermini versions of the Cray XMP, and selling them in locations were a full Cray was not needed. Cray purchased this company, and used them to develop a full line of lower cost Cray computers. This line consists of the following machines, when a larger PVP version exists, it is noted in parenthesis; XMS (XMP), EL/EL-92/EL-94/EL-98 (YMP), J90, J90se, SV1. Unlike the vector-MPP conflict, the lineage of the superminis is such that they are considered full Cray’s. Finally, there exists the main CRI line of computers, sometimes referred to as the PVP systems, this set is made up of these systems; Cray 1s/Cray 1m, XMP, YMP, C90, T90, X1. A few Cray branded machines exist outside of these classifications, such as the MTA (a Tera system, Tera now owns Cray, and uses the name), and the Cray-2 (a CCC machine).
In 1985, Seymour Cray began work on the Cray-3, under CCC. The Cray-3, and later the Cray-4 were unique in that they were both designed to be built using gallium arsenide (GaAs) based boards, and dissipated their 88kw of heat using Fluorinert (which was not that unusual for that vintage). CCC shipped one Cray-3 to the National Center for Atmospheric Research, I know of no Cray-4 installations, but CCC had Cray-4 boards running in Cray-3 enclosures when they closed. To the best of my knowledge, the Cray-4 was the final machine built by Seymour Cray before his accident.