Multithreaded and multi-core CPUs

From instruction parallelism to thread parallelism

Until the early 2000s, performance increases in single-processor CPUs resulted from higher clock frequencies on the one hand, and the use of instruction parallelism on the other. The article [H 1 058] describes these developments in detail. The "wall of heat" did not allow clock frequencies to be increased beyond 4 GHz. The use of instruction parallelism in superscalar processors with instruction execution in order [H 1 010] and not in order [H 1 011] has also reached its limits, although the use of data parallelism with SIMD instructions [H 1 200] has increased performance. To further improve performance, the use of thread parallelism in multithreaded and multicore processors has made it possible to combine all three types of parallelism: instruction parallelism, data parallelism via SIMD instructions and thread parallelism. The latter involves the use of parallel programming. This article introduces hardware support for multithreading and multicore processors. Most general-purpose processors in use in 2017 (PCs and servers) are multi-core processors with multi-threaded cores.