Computing is dead, long live supercomputing

International Supercomputing Conference 2007

The foundations of computer science and engineering must be shaken and reinvented to deal with the mass-market adoption of multi-core processors, Microsoft technical fellow Dr Burton Smith argued today. 

In his keynote address at the International Supercomputing Conference in Dresden, Smith told attendees that the " inflection point" of general-purpose parallel computing has been reached. 

"Our industry and the universities must work together to reinvent not only computing, but the computing profession," said Smith.

"The coming years will fundamentally reshape software and transform the way people use and interact with computers.

"In order for consumers to enjoy performance improvements in the future, mass-market technology providers will have to embrace parallel computing to differentiate and compete. It is vital that software and hardware adapt to new models of computing."

Smith urged commercial vendors to work with the academic and scientific communities to spur the next wave of discovery by creating software, tools and standards to help overcome existing barriers to parallel computing.

As uni-processor performance levels off, Smith predicted that three major problems will arise: instruction-level parallelism is near its limit (the ILP Wall); power per chip is getting painfully high (the Power Wall); and caches show diminishing returns (the Memory Wall).

"Microprocessors are now multi-core and/or multithreaded. But so far, it is just more of the same architecturally," said Smith.

Because of these constraints parallel computing is necessary for higher performance to get around these barriers and continue to offer better and faster computers.

But parallel programming is currently more complicated than serial programming as developers need to account for data and instructions that may be scattered over a variety of processors.

Smith urged the industry to work together to develop paradigms that make parallel programming for everyone who writes programs successful.

The good news, according to Smith, is that "thanks to HPC there's been a lot of pioneering and we're not starting from zero".

"Those of us who invented programming are still alive. We did it once can do it again," he said.