The National Security Agency's proposed Supercomputer Research Center in Prince George's County is a strong signal indicating how vital computing power has become to the defense establishment, several national security sources say.

The top-secret NSA, far more than its better-known counterpart, the Central Intelligence Agency, is responsible for intercepting foreign communications and decoding them so they can be analyzed for intelligence and foreign policy purposes. Similarly, the agency plays a key role in keeping sensitive U.S. telecommunications secure and encrypted to such a degree that decoding them is practically impossible.

Developing the sophisticated techniques necessary to protect American communications and crack the communications of others has prompted an enormous investment in state-of-the-art computing power. The faster and more powerful a computer is, the more likely it will be able to "cycle through" the possible mathematical sequences to decode an encrypted message.

Consequently, the NSA is one of the world's leading users of supercomputers, although the number of such machines now in use at the agency is classified information.

Supercomputers are machines capable of calculating at speeds thousands of times faster than conventional mainframe computers. Speeds are measured in "megaflops" -- for millions of floating point operations per second -- and "gigaflops" -- billions of floating point operations per second. Partly because of their cost -- usually between $5 million and $15 million each -- there are less than 100 such machines installed worldwide, according to High Technology magazine.

Besides underscoring how critical that technology is to defense, development of the supercomputer facility in Maryland is also a sign that the shadowy world of high-speed electronic data transmissions and sophisticated cryptoanalysis techniques has changed to the extent that the NSA no longer can rely on traditional alliances with the university and national laboratory communities to meet its technological needs in intercepting signals and breaking codes.

"We have found over the years that our needs are such that we have to engage in considerable research and development ourselves," said Mike Levin, an NSA spokesman. "We already have an extensive R&D operation, but this development grew out of a feeling that we had to develop a greater capability than currently exists."

Most existing supercomputers rely on a fairly linear architectural design. They process data as if it were pouring down a single pipeline. What the proposed supercomputer research facility is designed to do is explore whether "parallel" supercomputer architecture, with data pouring through several pipelines simultaneously, may be more effective in breaking ciphers faster.

One defense official notes that the real problem may not be the supercomputer architecture but whether encrypted messages lend themselves to parallel processing decryption techniques. This may be one of the key questions facing the NSA over the next decade.

One former government official, who asked not to be named, asserted that the NSA has done projections for its computation needs and concluded that it has to explore the potentially higher-speed parallel architectures to keep up with increasingly complex mathematical techniques used in encryption.

"Our research will be directed to the fields of cryptography," said NSA's Levin, "and the kind of thing we are aiming for does not exist yet."

The research facility, which will operate on a classified basis, complements the Defense Department's Strategic Computing initiative, which focuses more on artificial intelligence software techniques.

However, both the Defense Advanced Research Projects Agency (DARPA), which is coordinating the strategic computing program, and the NSA say they will cooperate in exchanging material on parallel computing architectures for supercomputers.

"We're going to work with DARPA," said Levin. "But this is a major effort that would be more concentrated than DARPA would have been."

The NSA initiative is believed to have come from a recommendation from the White House's Foreign Intelligence Advisory Board asserting that the U.S. intelligence community had to bolster its computational resources.