IBM’s $100 million bid to build a supercomputer that’s 500 times more powerful than the fastest computers today will be a challenge to pull off in the promised five years.
But it’s by no means impossible, said Arvind, a professor of computer science and technology at MIT.
“It is going to require brilliant engineering but I think it is possible to do it. They have the capability, and their architecture makes a lot of sense,” Arvind said.
IBM last week launched an ambitious research project to build a computer that will be capable of calculating 1 quadrillion operations per second, or 1 petaflop. The performance would make it about 1,000 times faster than IBM’s chess-playing Deep Blue supercomputer and about 2 million times faster than today’s most powerful PCs.
Nicknamed Blue Gene, the system will initially be used to perform a transaction that will take a full year — but that would take 1,000 years with today’s supercomputers.
The transaction will model the folding of human proteins, a process in which the strings of amino acids that form a protein fold into highly complex 3-D shapes that determine the protein’s function.
This folding process will involve more than 1 trillion calculations for each of the 1 billion ways that a protein can fold.
The model that IBM will initially run on Blue Gene — involving a 324-amino-acid protein — will take one full year to compute.
“No one’s been able to do this before,” said Bernard R. Brooks, a principal investigator at the National Institutes of Health in Bethesda, Md. “It could be done in principle, but today’s computers are simply not fast enough.” Today’s fastest computers would take more than 1,000 years to do the same thing, said Monty M. Denneau, an IBM research staff member.
A better understanding of the protein-folding process would provide better information on diseases and ways to combat them, said Ambuj Goyal, vice president of IBM’s systems and software group.
The tennis-court-size Blue Gene will have 1 million processors, each capable of executing 1 billion operations per second. The system will be based on a highly optimized massively parallel architecture capable of automatically detecting and overcoming failures of individual processors and computing threads. IBM calls the architecture SMASH, for Simple, Many And Self-Healing.