Abstract: GPUs have evolved from traditional graphics accelerators into core compute engines for a broad class of general-purpose applications. However, current commercial offerings fall short of the great potential of GPUs largely because they cannot be managed as easily as the CPU. The enormous amount of hardware resources are often greatly underutilized as there is hardly effective architectural support for fully managing them even when the GPU is shared across multiple applications.

To make GPUs a first-class controllable resource, we developed new architecture features to enable fine-grained sharing of GPUs, termed Simultaneous Multi-kernel (SMK), in a similar way the CPU achieves sharing via simultaneous multithreading (SMT). With SMK, different applications can co-exist in every streaming multiprocessor of a GPU. High resource utilization can be achieved by exploiting heterogeneity of different application behaviors. Resource apportion among sharers are developed for fairness, throughput, quality-of-services, etc. We also envision that SMK can enable better manageability of GPUs and new features such as more efficient synchronization mechanisms within an application.

Short Bio: Jun Yang is an Associate Professor of Electrical and Computer Engineering Department at the University of Pittsburgh. She received her Ph.D. from The University of Arizona, 2002, and became an assistant professor at UC Riverside (2002-2006) prior to joining University of Pittsburgh. Jun’s research is in the broad area of computer architecture and her recent focuses include GPU designs, emerging memory technologies, interconnection networks, 3D integration, and power and thermal management techniques. Jun is a recipient of NSF CAREER award in 2008, IEEE MICRO Top Picks award in 2010, and best paper awards of ISLPED 2013 and ICCD 2007.