Multicore processors are now commodity items, and this has created an unprecedented buzz about exploiting parallelism to maximize performance. This publicity has renewed interest in a long-standing problem: how much parallelism can we really exploit? Can extreme parallel computing be successfully delivered to the masses?
Several architectures are poised to expoit parallelism to achieve orders-of-magnitude speedup, including FPGAs, GPUs, Cell, and Manycore processors. Which of these architectures is the best approach? Or is there another stealth architecture that will be better? For a given application, how does one decide which approach is best?
All of these approaches work extremely well in their intended application domain. However, they all strive to become more general-purpose in nature. Will only one or two work out in the long run? Or, is a marriage in the cards for several of these architectures? What are the main obstacles standing in their path?
| 1. | Martin Langhammer, Chief Scientist, Altera | ||
| 2. | Dr. David Kirk, Chief Scientist, NVIDIA | ||
| 3. | Mike Butts, Fellow, Ambric | ||
| 4. | Dr. Fabrizio Petrini, IBM TJ Watson Research Center | ||
| 5. | Prof. Thomas Sterling, Louisiana State University | ||
| Guy Lemieux University of British Columbia | Tarek El-Ghazawi The George Washington University |
Categories and Subject Descriptors
C.1.4 [Processor Architectures]: Parallel Architectures; C.5.4 [Computer System Implementation]: VLSI Systems
General Terms
Algorithms, Design, Performance
Keywords
Custom compute engine, FPGA, parallel processing, reconfigurable computing
Copyright is held by the author/owner(s).
FPGA'08, February 24-26, 2008, Monterey, California, USA.
ACM 978-1-59593-934-0/08/02.