US nuke boffins say they have seen the future of multicore computing, and it is troubled.
Researchers at the Sandia national lab say that their projections indicate that performance gains flatten out badly after quad cores and cease altogether after eight - and beyond that point, performance actually worsens as more cores are packed onto a processor chip.
According to the Sandia boffins, the issue is that as numbers of cores increase, access to the information to be processed becomes more difficult. Performance gains extrapolated from Moore's famous Law can't be sustained.
"Multicore gives chip manufacturers something to do with the extra transistors successfully predicted by Moore's Law," says Sandia's Arun Rodrigues. "The bottleneck now is getting the data off the chip to or from memory or the network."
Rodrigues and his colleagues ran simulations which indicated that a 16-core unit would actually perform "barely as well" as a dual-core one.
"To some extent, it is pointing out the obvious — many of our applications have been memory-bandwidth-limited even on a single core," says Rodrigues. "However, it is not an issue to which industry has a known solution, and the problem is often ignored."
The Sandia boffins say that multicore systems have successfully taken supercomputing into the petaflop (a quadrillion floating-point operations per second) era, but they aren't going to break the exaflop barrier without something new.
"The [chip design] community didn't go with multicores because they were without flaw," says Mike Heroux, another Sandia computing brainbox. "The community couldn't see a better approach. It was desperate. Presently we are seeing memory system designs that provide a dramatic improvement over what was available 12 months ago, but the fundamental problem still exists."
The Sandia lab is partnered with the Oak Ridge lab, home to the world-number-two-ranked "Jaguar" petaflop machine, on an exaflop push called the Institute for Advanced Architectures. The Sandia boffins say that the Jaguar is based on their Red Storm design, and that they have a "large investment in message-passing programs" which "may help solve the multicore dilemma".
The US Department of Energy, which operates the Sandia, Oak Ridge and other American nuke labs, has traditionally been interested in supercomputing in order to simulate atomic warhead performance. The idea is that very accurate sims using huge amounts of computing power will allow the US nuke arsenal to be maintained in reliable condition without the use of live tests.
There's more on the multicore research from Sandia here. ®