The Impact of Multicore on Math Software
Power consumption and heat dissipation issues are pushing the microprocessors industry towards multicore design patterns. Given the cubic dependence between core frequency and power consumption, multicore technologies leverage the idea that doubling the number of cores and halving the cores frequency gives roughly the same performance reducing the power consumption by a factor of four. With the number of cores on multicore chips expected to reach tens in a few years, efficient implementations of numerical libraries using shared memory programming models is of high interest. The current message passing paradigm used in ScaLAPACK and elsewhere introduces unnecessary memory overhead and memory copy operations, which degrade performance, along with the making it harder to schedule operations that could be done in parallel. Limiting the use of shared memory to fork-join parallelism (perhaps with OpenMP) or to its use within the BLAS does not address all these issues.
KeywordsDistribute Memory System Core Frequency Panel Factorization Dense Linear Algebra Math Software
Unable to display preview. Download preview PDF.
- 2.Kurzak, J., Dongarra, J.J.: Implementation of Linear Algebra Routines with Lookahead - LU, Cholesky, QR. In: Workshop on State-of-the-Art in Scientific and Parallel Computing, June, 2006, Umea, Sweden (2006)Google Scholar
- 4.Sutter, H.: The Free Lunch Is Over: A Fundamental Turn Toward Concurrency in Software. Dr. Dobb’s Journal 30(3) (March 2005)Google Scholar
- 5.Asanovic, K., et al.: The Landscape of Parallel Computing Research: A View from Berkeley, Electrical Engineering and Computer Sciences, University of California at Berkeley, Technical Report No. UCB/EECS-2006-183 (December 18, 2006), http://www.eecs.berkeley.edu/Pubs/TechRpts/2006/EECS-2006-183.html