It’s beginning to look as if that alleged illicit sale of the design secrets of Saxpy Computer Corp, Sunnyvale, California to the Soviet Union (CI No 797) was a very serious leakage of technology indeed. According to Microbytes, many technologists were at first sceptical of the government charges, believing that the incident was just another attention-grabbing ploy on the part of the G-men. But as details about Saxpy and its Matrix 1 supercomputer architecture became more widely known, the sceptics began to express concern. Maybe the Russians really did get something after all, one engineer said to Microbytes Daily. The key point about the Matrix 1, and the one that makes it so attractive to the Soviet computer industry, which is still stuck somewhere back in the late 1970s, is that it was conceived to deliver tomorrow’s performance from yesterday’s technology. We make no bones about it, Saxpy vice president Joe Straub told the newswire. Our hardware technology is old and slow. It is really amazing that we are able to attain such speeds with relatively slow technology. Instead of putting effort into the design of expensive leading edge chips, Saxpy concentrated on developing advanced algorithms that enable performance of up to 1 GFLOPS to be squeezed out of off-the-shelf componentry. Saxpy claims that the Matrix 1 can match the speed of the Cray-2 and outperform the Cray X-MP/24 10-fold. Yet the hardware is based upon commercially available ECL and TTL components – bit-slice microprocessors, presumably – so that the cost is dramatically lower, enabling a Matrix 1 to sell for between $900,000 and $1.8m against $8m to $25m for a Cray 2. Moreover the tried and tested hardware technology enables lower power requirements and an air-cooled design. As the machine’s name implies, the algorithms developed for the Matrix 1 are conceived for efficient execution of matrix arithmetic, where the structured algorithms operate on large, multi-dimensional blocks of data. Up to 32 parallel pipelined processors are used to create a single-instruction, multiple-data architecture to work on up to six dimensions of a problem simultaneously. The processor can perform up to 32 floating-point multiplications and 32 floating-point additions or subtractions simultaneously. The system is controlled by a processor running the DEC VAX/VMX operating system and is programmable in Fortran-77, C, Pascal, and Ada. Applications so far have mainly been in scientific and engineering signal processing work – the US government uses the Matrix 1 for simultaneous processing of several radar signals so as to determine the speed, location, and size of an object more precisely. And the technology that makes that possible may well be what Soviet scientists now have to play with.
Content from our partners
