NASA Jet Propulsion Laboratory, California Institute of Technology

NASA's Jet Propulsion Laboratory has announced the inauguration of its Mark III Hypercube parallel supercomputer.

The Mark III is the result of five-year research and development effort at the JPL Center for Space Microelectronics Technology in collaboration with Dr. Geoffrey Fox of the California Institute of Technology.

It represents the arrival of massively parallel supercomputing.

The first module of the Mark III, which was placed on-line on the Caltech computer network July 1, contains 32 nodes, or processing units, which together have peak speed of about 512 million floating point operations per second (flops).

Three more 32-node modules will be added during the next nine months to form 128-node hypercube with each single node having the power of 25 VAX minicomputers. The peak speed of the 128-node Mark III is 2 billion floating point operations per second. This performance makes it more powerful than conventional supercomputers such as the Cray 2.

In addition to breaking the speed barrier, the parallel hypercube is about 10 times more cost-effective than a conventional supercomputer because it is built with the same low-cost, mass-produced Motorola 68020 microprocessors and Weitek floating chips that are used in personal computers.

Moreover, the hypercube architecture is scalable so that 10-fold performance increase can be obtained by increasing the number of nodes by factor of 10. New massively parallel machines are envisioned that exceed today's performance by factor of 1,000.

The Mark III Hypercube is joint endeavor of Caltech and JPL scientists and engineers dating from 1983. It is based on the pioneering work of Caltech computer scientist Charles Seitz and physicist Geoffrey Fox who researched the hypercube type of parallel computer architecture in the late 1970s and began construction of laboratory prototype in 1980.

The Mark III is the third generation of hardware in the Caltech/JPL Hypercube development. Four commercial firms are now selling second-generation hypercube technology and there are about 100 hypercube installations worldwide.

The Mark III Hypercube is operated by the Caltech Concurrent Supercomputer facility, which is the first supercomputer facility dedicated to massively parallel machines. The facility also includes commercial parallel computers made by Intel, NCUBE, Thinking Machines and Ametek.

Parallel computers are solution to fundamental limit on the speed of single computers. Because of the finite speed of light, single processor is limited as to its ultimate speed.

Therefore, in order to achieve major speedup in problem solving, many single processors must be coordinated to work simultaneously on single large problem, just as groups of workers subdivide the work and coordinate their efforts to complete task too large for single worker.

To work efficiently on problem, parallel computer must communicate its intermediate results among the many individual processors. The layout of the communication between the processors determines the "architecture" of the computer. hypercube characteristically has 2, 4, 8, 16, etc., processors.

The Mark III Hypercube is being used for scientific, engineering and defense research applications such as:

- Study of the spin structure of liquid helium by Michael Cross of Caltech.

- Structure of quark-quark interactions in the proton by Geoffrey Fox of Caltech.

- Analysis of NASA multispectral space imaging data by JPL scientist Jerry Solomon.

- Simulation of the strategic defense system led by JPL's David Curkendall.

- Analysis of NASA synthetic aperture radar images taken from the space shuttle by JPL's Jean Patterson.

The Mark III Hypercube research and development is sponsored by the U.S. Air Force Electronic Systems Division, the Department of Energy and NASA.

News Media Contact