A professor at the University of Maryland's A. James Clark School of Engineering, armed with a grant from the Department of Defense and the National Science Foundation, figures he's close to having a general-purpose parallel processing chip that's easy-to-program and could, well, it could replace the Pentium in mainstream computing - except that it would be capable of computing speeds a hundred times faster than current desktops, accomplishing multiple tasks in parallel.

The trick is in the algorithms.

Professor Uzi Vishkin would be working at Intel's research operation at Carnegie Mellon in Pittsburgh but for a little difference of opinion over who should own the IP.

As it stands Professor Vishkin's Delaware-based company XMTT Inc owns the patents on the unnamed chip and he reckons he's a league ahead of Intel - at least in being ready to explain how parallel computing can be easy to program.

Easy enough that this fall he's going to have some high school kids try programming his 75MHz desktop eXplicit Multi-Threaded (XMT) prototype, 11 years in the making.

He said it can be done using a standard C programming language variant called XMT-C that adds only two commands: the key one is spawn, a way for programmers to basically conjure up an infinite number of virtual threads and let the machine take it from there.

The beauty of it is all those threads look like a single thread to the programmer.

What the professor and his colleagues have in hand right now is a prototype circuit board about the size of a license plate fitted with 64 Xilinx FPGA parallel processors. He figures they're about six month away from sending a single chip off to IBM to be fabricated.

He's aiming at a single part the size of a fingernail with 1,024 light slave cores surrounding a master core that should clock at 1GHz-2GHz. He imagines the clock could scale and match any cutting-edge gismo.

The Viskin chip is operating system-agnostic and would take some of the heat off the operating system people in coping with increasingly multi-threaded programs. The professor says he would have no problem making his approach x86-compatible.

The chip's architecture is what they call a Parallel Random-Access Machine or, in this case, PRAM-on-a-Chip, a concept that won the theory battle 20 years ago but couldn't be commercialized until chips got enough bandwidth, the professor says.

New York University's Ultracomputer, IBM's PR3, Saarbruken's PRAM machines are buried in the PRAM graveyard. Cray's large multi-chip MTA supercomputer effort soldiers on.

Vishkin says PRAM was ahead of its time. Now that a 30 billion-transistor budget is in sight for chips there's enough bandwidth to handle what's effectively a "superhighway" or interconnection network connecting scores of cores on a sliver of silicon.

Anyway, the prof claims, it's PRAM or nothing. There's no other feasible alternative and he's critical of what he calls the industry's "build-it-first figure-put-how-to-program-it-later" architectures.

Vishkin calls the industry "clueless," and quotes MIPS founder John Hennessy, now the president of Stanford University, recalling that "many of the early ideas were motivated by observations of what was easy to implement in the hardware rather than what was easy to use."

Vishkin says it too utterly painful to program in other parallel approaches.

Vishkin showed off the prototype at the ACM International Conference on Supercomputing (ICS) in Seattle earlier this month and conference-goers were able to connect to it remotely and run programs.

Last week the professor was at Microsoft's invitation-only Workshop on Many-Core Computing. Microsoft Technical Fellow Burton Smith figures the device can "exploit a wider spectrum of parallel algorithms than today's microprocessors can and this in turn will help bring general-purpose parallel computing closer to reality."

Smith was in Dresden at the International Supercomputing Conference the other day saying that computing has got to be reinvented so garden-variety programs execute in parallel in multiple microprocessor cores to pave the way for speech, conversation, rich visualization and anticipatory execution of tasks. He urged commercial vendors to work with the academe.

Wonder how Intel feels about that.