The Beast from Below: How Changes in the Hardware Ecosystem Will Disrupt Computer Science

Опубликовано 8 августа 2016, 19:02

For decades, computer scientists have relied on steadily advancing hardware capabilities. Moore�s Law coupled with Dennard Scaling have improved von Neumann computing by many orders of magnitude. Notably, the underlying changes have been largely invisible to software, with standard instruction sets and compilers hiding the complexity that has produced those enormous gains. However, the easy times are ending; many orders of magnitude gains in performance and efficiency are still achievable, but getting there will break our current notions of instruction sets, compilers, languages, data types, circuits, devices, and applications. Different classes of applications will be enabled, and whole new areas in computer science will emerge. Fortunately, many of the exciting emerging workloads (machine learning, computer vision, speech recognition, and so forth) will be amenable to this new era. In my talk, I will cover the reasons for this shift, and make some predictions about how it will affect our field.