Project: Evaluation of State-Of-The-Art Hybrid Hardware Architectures based on Application Bottlenecks

Team: Dr. Peter Tröger, Frank Feinbube

Research institution: Hasso-Plattner-Institut Potsdam

Abstract: Trends in hardware developments emphasize the every-increasing importance of hybrid computing for future computer architectures and software systems. Competitive applications need to leverage the performance opportunities provided by emerging generations of accelerator technology. With the introduction of its K20 architecture, NVIDIA takes a big step towards a wider applicability of GPU computing by supporting new concepts like dynamic programming, on-device grid management, and direct data exchange. Intel's novel Xeon Phi accelerator, being a stand-alone PCI-Express card like GPUs, but still being x86 compatible, is approaching the field of hybrid computing from the general purpose side. The purpose of this study is evaluate the feasibility of the K20 architecture and the Xeon Phi accelerators for well-known application bottlenecks.

Link to the previous project: public/20132030

Last modified 5 years ago Last modified on Apr 9, 2014 6:31:56 PM