Ever since last year, all the talk in the processor space has been about dual-core and multi-core. This new technique of putting more than one processor core on a single chip was presented as the answer to offering increased microprocessor performance.
Intel and AMD both said they were pursuing dual-core and multi-core chips to address the challenge of offering computing customers the performance increases they had grown accustomed to now that the technique of increasing frequency had hit a brick wall.
But at In-Stat’s Spring Processor Forum in the US this week, AMD seemed to pull back from that. Chuck Moore, senior fellow at AMD said in his keynote address that software to take advantage of the performance increases possible with more than one core on a chip was not here yet, which presented a highly complex problem that would not be fixed overnight.
“Single-threaded applications are not going away tomorrow,” said Moore. “We are ready for throughput-oriented workloads, but we are not ready for parallel-oriented workloads.”
Computer programming students have all been trained on single-threaded applications, and those programmers who have created applications for parallel-oriented workloads have worked on the problem for many years to achieve anything that comes close to delivering the promised performance without errors.
Such applications raise the level of complexity to a whole new level, in some cases causing a hit to performance because of it.
In the future, computer programming students will be trained in creating applications for parallel workloads, Moore said. But until that happens, getting the performance increases that users have become accustomed to out of dual-core and multi-core processor systems may be an elusive dream.