Do the Parallel Computing Toolbox and MATLAB Distributed Computing Server perform better on multi-core systems or multi-processor systems?

When using the Parallel Computing Toolbox and MATLAB Distributed Computing Server, is better performance offered when using a computer with a single processor and multiple cores, or multiple single-core processors?

Answer 1

[MathWorks Support Team]

As an organization, MathWorks is not able to provide hardware recommendation. We would advise you to consider some of the common trade-offs that are associated with multi-core and multi-processor systems.

For example, multi-processor systems are known to have an advantage over multi-core systems by virtue of each processor having its own front-side bus (FSB). On the other hand, multiple cores in a single processor share a common bus, thus limiting the communication bandwidth to the cores. However, when communication between multiple cores is enhanced because they reside on the same piece of silicon, this is known to mitigate the effects of their sharing a front-side bus.

Note that over the years, different processor and motherboard vendors have developed hardware architectures that compensate for these effects, consequently blurring the performance difference between multi-core and multi-processor systems at times. The vendors themselves may be able to provide additional insight in these respects.