Prior to the 1990s what we'd usually call "supercomputers" (IE, in the "Cray" sense) were usually assembled using parts designed strictly for the job, often fabricated using more exotic fabrication processes than those used in microcomputer-based hardware. And the architectures were usually geared toward achieving one (or a few) very fast execution cores; the mid 80s Cray-2 had "only" four vector cores, but in clock cycle times alone they ran about 20 times faster than the best microprocessors of the era could offer. (And the entire unit together could perform anywhere from a hundred to a thousand times faster than a typical desktop machine on jobs that were well suited to it.) Thus any "super" machine built out of 680x0 CPUs would by definition have to be a massively parallel unit and, well... the slowest/oldest unit in that class using commodity parts I've seen any references to is the
Intel iPSC "Hypercube", constructed out of anywhere from 32 to 128 80286+80287 CPU/FPU "nodes". (Many/most MPP machines, like the early Connection Machine CM-1/2s or the competing MasPar units used custom computing elements, not commodity CPUs.) Subsequent iPSC/Paragon machines were built by Intel using 80386s, i860s, and Pentium Pros, but it's debatable if any of the systems built prior to the 1993-era i860 Paragon XP/S really qualified as a
Supercomputer.
Are we talking about computers which were "super" compared to a typical desktop/minicomputer or genuine "fastest in the world" contenders? Cray's first "commodity CPU supercomputer" (if you include "workstation" CPUs in the commodity definition), the Cray T3D of 1993, was built on the DEC Alpha 21064 which is a ridiculously powerful chip compared to *any* member of the Motorola 680x0 family. (Easily seven to ten times as fast as the 68040 in a typical application.) Later model Connection Machines (and others) used SPARC and of course, SGI had MIPS-based systems that would qualify as
"super".
ASCI Red, the first supercomputer to crack the one teraflop mark back in 1997, was made out of Pentium II Xeons and advertised by Intel as " the first large scale supercomputer to be built entirely of common commercially available components". It certainly wouldn't surprise me to find out that an experimental 680x0-based massively parallel cluster existed at some point, probably in the form of racks of VMEbus boards filling a room in a college basement, but as a commercial product? Any evidence of that escapes me. If someone has a citation it might be interesting reading.
(There are examples of 680x0-based of computers acting as the front end for a cluster composed of other CPUs, such as i860s or Transputers, that are fairly easy to find. But that doesn't count.)
