Programming of shared memory GPUs shared memory systems
School of Information Technology and Engineering, University of Ottawa, Ottawa, Ontario
University of Ottawa, 2011
@article{bergeron2011programming,
title={Programming of shared memory GPUs shared memory systems},
author={Bergeron, J.P.},
year={2011}
}
The concept of shared memory is known to engineers for a long time: a large block of memory shared between processors to allow for synchronization. But what if there where thousands of threads with hundreds of processors accessing a single memory block. This challenge is discussed in this paper. There are APIs used to program the graphics processing unit (GPU) which takes advantage of the high number of processing elements. The memory architecture is specifically designed in order to obtain the maximum bandwidth. A great speedup is achieved after optimizations.
January 2, 2012 by hgpu
