Silexica: Mastering Multicore

By _Cadence - Breakfast Bytes

December 4, 2017

Since the invention of the microprocessor, it was a dream that it would be possible to build a really powerful computer by taking a lot of cheap simple computers and putting them together. This was especially a dream of hardware designers, who could see their way to addressing the hardware problems, and then the rest was "just" software. That software turned out to be difficult to create. About a decade ago, it became clear that microprocessor clock frequencies could no longer be increased and companies like Intel switched to multi-core processors. However, that software was still not easy to write, and it was hard to make use of large numbers of cores for individual jobs.

There were some tasks that work well on this sort of fabric. Some programs are "embarrassingly parallel" with almost limitless opportunities. The standard example is graphics where each pixel can sometimes be processed individually, with reference only to a few nearby pixels. In fact, it is this that allows GPUs to function, since even when dealing with polygons and shading, it is still the case that one part of the image is largely independent of parts that are not in the immediate neighborhood.

To read the full article, click here