Andrzej Nowak (CERN)
"Architectural Details and Performance Studies" Considering the rise of complex many-core processors, a sufficient understanding of their architecture and of the relevant performance tuning opportunities has become an indispensable element of software development. Although by using various tools we are often able to get a generous peek both inside the hardware and software, drawing high-level conclusions that impact our software is not always straightforward. Another considerable challenge comes from accelerators and co-processors, where different computing paradigms take precedence, such as extreme data parallelism, different math capabilities, the importance of special languages as well as memory size and topology changes. The objective of this lecture is to bring the audience closer to "where it matters" in modern computing systems, and to identify important paths for high performance.