By Matthew Dublin

In what could be construed as either self-effacement or some form of mild capitulation, multicoreinfo.com has posted a recent study that says multi-core processor technology is not the Charlie's golden ticket to the Willy Wonka performance factory we all thought it was. The report entitled "The Diminishing Returns of Moore's Law: Multicore Is Not the Panacea for Increasing Performance It Should Be" is published by IDC and included in marketresearch.com's collection of hardware reports. The report, available for the bargain-basement price of $3,500.00, presents IDC's take on all the hullabaloo surrounding the proliferation of multi-core processors and the impact they have — and do not have — on software performance. Like a bunch of teenagers who have just gotten busted for something naughty, the multi-core blogosphere seems to just be staring at their shoes in stony silence. Now, while we admittedly have not forked over the cash for the entire report, the big pull quote and de facto thesis statement says that the promise of multi-core has fallen short due to a "lack of application development tools to efficiently program applications for parallel execution." The folks over at Rapidmind and Tilera likely beg to differ. Intel also has some cool tools that let you dip your toes into parallelism before jumping in headfirst. If you would like to simulate what your serial applications would do in a parallel parallel universe (sorry, couldn't help myself), check out Intel Parallel Advisor Lite. Cray is also working on developing Chapel, an open source HPC programming language for multi-core and parallel systems, and the list goes on.

But to review for those of you not versed in the proper background "geekery," here's the very long story shortened: Moore's law states that every two years the number of transistors you can cram onto an integrated circuit increases exponentially, which is why by the time you paid for your new computer, it was already outdated. But by 2005 this "law" hit a wall and the free lunch was over for developers — things just couldn't get any faster. Enter the age of multi-core processors. More processing cores on a chip means that bioinformatics algorithms or programs like R, VMD, Charm++, and NAMD have increased performance, if properly tweaked. And sure, it is not always a walk in the park - in some cases, getting your applications to run on multi-core architecture requires a somewhat intimate knowledge of the serial code - but it hardly means that multi-core has failed to deliver. Breeding grounds for the next generation of programmers like Georgia Tech and MIT have made multi-core programming a key part of the curriculum, and the code tinkerers with chutzpah enough to venture into this territory have already laid a lot of ground work. Cognitive Consonance has an interesting post from earlier this month that takes apart these different philosophical camps that exist amongst bioinformatics folks trying to toy with multi-core.The lower level C/C++/Fortran crowd seems to be one side and MPI/Python people on the other. The discussion is happening, the SDKs exist and work, everybody just needs to lighten up.