Tuesday, February 25, 2014

This project aims in the design of a novel multicore 3D Graphics Processing Unit (Think3D) and its associated software that would allow the efficient rendering of multimedia content on mobile platforms under tight power constraints. Think3D focuses on the acceleration of graphics applications based on the OpenGL|ES 3.0-Khronos standard, on video decoding according to H.264/MPEG4/VC7 standards and on supporting general processing workloads via the OpenCL programming model that forms the base for future standards focusing on embedded computer vision and augmented reality applications. Think3D is built using the Nema ultra-threaded processor (up to 128 simultaneous threads) designed by Think-Silicon that offers low cost implementations as enabled by the adopted power-conscious hardware-software co-design methodology. Scaling Think3D to a large number of Nema cores (64/128+) requires also a scalable on-chip communication fabric that would connect efficiently the processors and the on-chip memory system. Additional communication-related performance improvements are expected by the novel memory management methodology that removes redundant traffic thus saving both precious on-chip bandwidth and power. Agility and flexibility are required to differentiate companies from the competition because time to market combined with the right and unique features are often the critical factor that determines success. Think3D platform is a novel IP subsystem that addresses the needs of multiple closely related markets (3D graphics, video decoding, computer vision and augmented reality) since it provides the necessary computation capabilities and the needed programmability to cover many application domains with the same hardware platform. This reusability and programmability property offers three distinct advantages: (a) reduced time to market allowed by simple domain-specific customizations of the Think3D platform (b) amortized per-application design cost (c) place Think-Silicon in head-start of the emerging era of heterogeneous computing that is expected to dominate the embedded systems market.