As a researcher involved in parallel processing and parallel computing, this article has really crystallised some of my thoughts about the field.
I sometimes hesitate to even call our current multi- and manycores parallel processors. The actual superscalar architecture is actually very parallel, you just have direct control over any of the parallel features baked into the silicon.
You have market forces limiting the exploration of vastly different architectures. Our multicore processors are the result of small incremental changes to the efficient sequential microprocessor design that is the von Neumann architecture. Even if a designer bakes a parallel processing feature, such as the 'micro-dataflow' out-of-order execution cores; this still needs to fit into the high-level processor architecture. In other words, the processor still needs to pretend it is a sequential one.
1 comment
[ 3.2 ms ] story [ 20.8 ms ] threadI sometimes hesitate to even call our current multi- and manycores parallel processors. The actual superscalar architecture is actually very parallel, you just have direct control over any of the parallel features baked into the silicon.
You have market forces limiting the exploration of vastly different architectures. Our multicore processors are the result of small incremental changes to the efficient sequential microprocessor design that is the von Neumann architecture. Even if a designer bakes a parallel processing feature, such as the 'micro-dataflow' out-of-order execution cores; this still needs to fit into the high-level processor architecture. In other words, the processor still needs to pretend it is a sequential one.