During Intel CTO Justin Ratner’s keynote today at IDF 2011, it was revealed that the company is prepared to release an entirely new kind of CPU – this time, it has more than fifty cores. Coming from the Knights platform of research within the company, the project is codenamed “Knights Corner”.
As you can see in the picture above, the Knights Corner platform is chock full o’ cores. Where Knights Corner chips can excel is in the processing of heavily parallelizable workloads. It’s almost a bridge between the general-purpose CPU and the highly specialized GPU. When a program or task is parallelizable, it means that it can be broken up into discrete chunks and processed simultaneously, rather than sequentially.
Imagine you’re tasked with having to make rolls for dinner. If you worked sequentially, you’d have to shape one roll at a time, working through all of the dough until it had all been processed. If your family helps, however, you could work in parallel, with each of you taking a portion of the dough and rolling out your bread. That’s a simplifed version of the difference between parallel and sequential computing.
Some tasks simply can’t be processed in parallel. If you’re working on a problem that requires you to compute an equation, say, then perform some other computation on the first result &emdash; well, that couldn’t be broken up piecemeal. Applications, however, are growing increasingly capable of taking advantage of multi-core systems, and some of the most parallelizable software is that used in HPC, or high-performance computing.
On stage, Ratner brought up one of the principal researchers at the Large Hadron Collider (LHC) at Cern, in Switzerland. The data collected in the massive particle collisions requires substantive computer hardware in order to successfully analyze everything that the sensors gather. Other HPC uses include weather modelling, stellar astronomy analysis, physics problems, protein modelling and more.