Janos Vegh

Can Broken Multicore Hardware be Mended?

The multi- and many-core (MC) era we have reached was triggered after the beginning of the century by the stalling of single-processor performance. Technology allowed more transistors to be placed on a die, but they could not reasonably be utilized to increase single-processor performance. Predictions about the number of cores has only partly been fulfilled: today’s processors have dozens rather than the predicted hundreds of cores (although the Chinese supercomputer [3] announced in the middle of 2016 comprises 260 cores on a die). Despite this, the big players are optimistic. They expect that Moore-law persists, though based on presently unknown technologies. The effect of the stalled clock frequency is mitigated, and it is even predicted [6] that ”Now that there are multicore processors, there is no reason why computers shouldn’t begin to work faster, whether due to higher frequency or because of parallel task execution. And with parallel task execution it provides even greater functionality and flexibility! ”