Programmable acceleration of data processing intensive workloads
Michael Gschwind (IBM T.J. Watson Research Center, USA)
The data collected by devices ranging from environmental sensors to medical equipment and consumer electronics is burgeoning. To increase the value of these increasingly large amounts of digital information collected from a variety of data sources, processing of large amounts of information is necessary. To meet the demand for increased processing power, accelerators for compute intensive processing offer increased area and energy efficiency by reducing core complexity and increasing core count. Accelerators achieve significant performance benefits over traditional microprocessor cores by giving programmers explicit control to optimize for efficient use of the off-chip bandwidth and latency.
The Cell/BE was the first system architecture integrating accelerators to accelerator processing-intensive tasks. In new Cell/BE architecture, programmers can exceed the performance obtained from using history-based caches by exploiting application knowledge about data usage patterns with compute/transfer parallelism and control over data placement, Today, the world.s fastest computer is a hybrid design exploiting accelerators for data processing intensive workloads based on the Cell/BE.
Increasingly, GPU vendors are also marketing their graphics processing units as numeric accelerators for these workloads. We will explore system architecture and programming model considerations for GPUs and Cell for use in accelerating data processing intensive workloads.
Dr. Michael Gschwind is the Manager of Architecture for IBM Systems and Technology Group where his team is responsible for future Mainframe, PowerPC and I/O architecture. He was one of the initiators and a leading contributor to the Cell Broadband Engine system architecture definition as well as a lead architect of the Synergistic Processor architecture. During the definition of the Synergistic Processor architecture, Dr. Gschwind also developed the first Cell Broadband Engine compiler and served as first technical lead for system software development. Dr. Gschwind was also a lead architect for the VMX128 SIMD extensions in the Microsoft Xbox360 system, and was a key architect and design lead for several technologies to be introduced in Power7. Most recently, Dr. Gschwind was the Blue Gene Floating Point Lead and Chief Architect of the QPU quad-processing unit for the next generation BlueGene/Q system. He has been recognized by IBM with the title Master Inventor and was recognized in 2006 as IT Innovator and Influencer by ComputerWeek. Dr Gschwind is a Fellow of the IEEE.