About processors, chips, sockets, and cores

Following are brief definitions for common terms related to supercomputers:

Recent developments in computational architecture can lead to confusion concerning what a microprocessor is. Since the advent of multi-core technology, such as dual-cores and quad-cores, the term "processor" has been used to describe a logical execution unit or a physical chip. A multi-core chip may have several cores. With the advent of multi-core technology, the term "processor" has become context-sensitive, and is largely ambiguous when describing large multi-core systems. Essentially a core comprises a logical execution unit containing an L1 cache and functional units. Cores are able to independently execute programs or threads. Supercomputers are listed as having thousands of cores.

A chip refers to a physical integrated circuit (IC) on a computer. A chip in the context of this document refers to an execution unit that can be single- or multi-core technology.

The socket refers to a physical connector on a computer motherboard that accepts a single physical chip. Many motherboards can have multiple sockets that can in turn accept multi-core chips.

A process is an independent program running on a computer. A process has a full stack of memory associated for its own use, and does not depend on another process for execution. MPI (Message Passing Interface) processes are true processes because they can run on independent machines or the same machine.

A thread is essentially a process that does not have a full stack of memory associated for it. The thread is tied to a parent process, and is merely an offshoot of execution. Typically thread processes must run on the same computer, but can execute simultaneously on separate cores of the same node. OpenMP parallelism uses threads for child processes.

Hyper-threading is an Intel technology that originally preceded multi-core systems, and was used to make a single core appear logically as multiple cores on the same chip. Intel abandoned hyper-threading briefly during the advent of multi-core processors but reintroduced the technology in 2008. Since then, Intel has used it extensively to improve the performance of parallel computations in its multi-core processors. Hyper-threading improves performance by sharing the computational workload between multiple cores whenever possible, allowing the operating system to schedule more than one process at a time. For more, see Intel Hyper-Threading Technology.

Multi-core compute nodes can be described by the number of execution units, or cores. A computer with 8 cores would be described as an 8-way node. This machine can have 8 independent processes running simultaneously. A 32-core system would be called a 32-way node.

As explained above, a processor could describe either a single execution core or a single physical multi-core chip. The context of use will define the meaning of the term.

This is document avfb in the Knowledge Base.
Last modified on 2019-02-15 09:53:51.