For many decades, software developers benefitted from being able to use the same software code while working with increasingly powerful hardware. As hardware manufacturers regularly improved the performance of their semiconductors by enhancing transistor densities and clock speeds, software development enjoyed a “free ride”—they were able to readily develop on these new devices without having to change software architectures.
However, processing power has hit a wall due to the limit in increasing clock speeds. As a result, chip manufacturers have been turning to dramatically new approaches to achieve further performance gains. First it was hyper-threading and homogeneous architectures, and then heterogeneous multicore architectures. To benefit from these hardware changes, existing software had to be parallelized and modified to deal with the heterogeneity.