PANEL: Data Centric Computing, Present and Future скачать в хорошем качестве

PANEL: Data Centric Computing, Present and Future

Presented by Hoshik Kim (SK hynix) | Ada Gavrilovska (Georgia Tech) | Pankaj Mehra (Elephance Memory) | Christos Kozyrakis (Stanford) The advancements in CPU performance through Dennard scaling and Moore's Law have encountered limitations, prompting the exploration of alternatives to the traditional CPU-centric computing model. Domain-specific computing, initially observed in graphics and network cards, has evolved into various accelerators like GPGPUs, TPUs, FPGAs, Computational Storage (CS), and Processing-In-Memory (PIM) among others. This shift towards domain-specific architectures aims to overcome these limitations. Notably, data-centric approaches, exemplified by big data processing and recent machine learning, have gained prominence in enhancing overall computing efficiency and performance. The OCP FTI (Future Technology Initiative) Data-centric Computing workstream is dedicated to addressing these challenges especially from the technology dissemination perspective. A focal point of the panel discussion in the breakout session is the exploration of computing and programming models pertinent to data-centric computing utilizing domain-specific architectures. Distinguished experts from academia and industry, well-versed in these matters over many years, have been invited to share their insights on the challenges ahead and right technology direction.