A new courseware page — topics in computer architecture — is now available. This page covers a wide mix of topics from RISC vs. CISC to non-von Neumann languages and architectures. You might just find an interesting lecture or two for a special topics seminar about computer architecture.

Way back when in the early eighties, I had a chance to work with Al Davis at the University of Utah. Al was a proponent of data-driven dataflow computers. His ideas showed me that there are alternatives to centralized synchronous control, sequential programming and the von Neumann memory bottleneck. The topics page has links to information about some of this interesting work.

Looking back, I’m struck at how mired we are in the von Neumann architecture! Sure, there are a few projects underway to investigate alternatives like quantum computing, functional programming languages, and so forth. However, the world remains dominated by the centralized, sequential von Neumann model and programming languages in which parallelism is glued on through a (class) library or two. General purpose GPU (GPGPU) computation is not really that radical. Yes, GPGPU computation is data parallel, but the underlying model is the sequential execution of SIMD instructions. Speculative execution is a bit of a kludge to allow conditional execution in highly sequential SIMD programs. Further, data transfer from memory to execution units remains a limiting factor in performance.

I think we need to introduce students to a wide range of computing models. Hopefully, some young, clever soul will find a way to pull us out of our rut!