A fully 80X87-compatible numeric co-processor claimed to offer up to 10 times the performance of the standard Intel Corp parts has been released by the Cyrix Corp of Richardson, Texas. The FasMath processor offers a peak performance of 5.5 MFLOPS by implementing floating point primitive operations in hardware rather than in microprogrammed sequencer. Cyrix says that this enables the processors to perform floating point operations as fast as an 80386 can perform integer additions. It implements a full extended double precision IEEE-754-1985 architecture using 80-bit internal format for storage and computation. The instructions used to program the processor are binary- and function-compatible with those defined for the Intel numeric processor, and 117 floating point instructions are implemented. Operating at 20MHz, 25MHz, or 33MHz, it dissipates 20W of power, said to be 5% lower than other processors. Cyrix says that the FasMath is the first in its family of math processors for workstations and personal computers that will enable desktop systems to run applications previously confined to mainframes. Cyrix has appointed a network of distributors in France, Germany, Switzerland, and the UK. The French distributor is Newtek of Cedex, and Atlantik Electronik GmbH is based in Martinsried in Germany. Switzerland is covered by Chiptec AG of Langenthal, and the UK distributor is the Thame, Oxfordshire company, Microcall Ltd. The 20MHz, 25MHz, and 33MHz processors cost $647, $814, and $994 respectively.
