Matches in SemOpenAlex for { <https://semopenalex.org/work/W636812313> ?p ?o ?g. }
Showing items 1 to 69 of
69
with 100 items per page.
- W636812313 abstract "In this thesis, the performance and energy efficiency of two current hardware platforms are evaluated, the Intel Sandy Bridge Core i7 and ARM Cortex-A9 MPCore test chip, using techniques like vectorization and multi-threading with task-based programming using OmpSs. We present results from three task-based programs, Black-Scholes, FFTW and matrix multiplication on both platforms. The performance and energy efficiency is compared between different configurations of threads, vectorization and task scheduling algorithms. Energy consumption is measured using the Running Average Power Limit interface on the Sandy Bridge, and regular sampling of the current power dissipation from the system configuration registers on ARM. The energy efficiency results are presented using the metrics total energy consumed, power, GFLOPS/W, and the energy-delay and energy-delay squared products. The energy efficiency of the ARM Cortex-A9 MPCore is compared to that of Sandy Bridge using the process-normalized energy-delay and energy-delay squared products, as well as GFLOPS/W and energy. Black-Scholes is adapted to use vector code, and FFTW is compiled with and without vector support. The matrix multiplication application uses ATLAS, which already is vectorized. Code for sampling and numerically integrating the power dissipation over time was developed for ARM, then different task scheduling algorithms are explored for each application. For both platforms, vectorization with SSE/AVX and NEON is found to consume little to no extra energy per second while giving significantly higher performance. Multi-threading gives higher performance, but with higher power consumption. With AVX on Intel, Black-Scholes shows an energy efficiency of 0.82 GFLOPS/W, FFTW up to 1.4 GFLOPS/W, and matrix multiplication almost 2 GFLOPS/W. Both multi-threading and vectorization significantly reduced the energy-delay products, showing up to 99.55% reduction for Black-Scholes and 93.65% for FFTW compared to non-vectorized single-threaded code for the Intel platform. The ARM results are restricted from publishing, but can be found in appendix A for those with access. Single-threaded execution is shown to give a better GFLOPS/W for small problem sizes in the benchmarked applications, while the EDP is reduced for multi-threading even for small problems. Black-Scholes show increased performance and energy efficiency with hyper-threading. FFTW shows no significant difference using hyper-threads, and matrix multiplication shows a negative effect on both performance and energy efficiency." @default.
- W636812313 created "2016-06-24" @default.
- W636812313 creator A5064912974 @default.
- W636812313 date "2012-01-01" @default.
- W636812313 modified "2023-09-27" @default.
- W636812313 title "Case Studies in Multi-core Energy Efficiency of Task Based Programs" @default.
- W636812313 hasPublicationYear "2012" @default.
- W636812313 type Work @default.
- W636812313 sameAs 636812313 @default.
- W636812313 citedByCount "0" @default.
- W636812313 crossrefType "journal-article" @default.
- W636812313 hasAuthorship W636812313A5064912974 @default.
- W636812313 hasConcept C119599485 @default.
- W636812313 hasConcept C126255220 @default.
- W636812313 hasConcept C127413603 @default.
- W636812313 hasConcept C173608175 @default.
- W636812313 hasConcept C206729178 @default.
- W636812313 hasConcept C2524010 @default.
- W636812313 hasConcept C2742236 @default.
- W636812313 hasConcept C2780165032 @default.
- W636812313 hasConcept C32900221 @default.
- W636812313 hasConcept C33923547 @default.
- W636812313 hasConcept C3826847 @default.
- W636812313 hasConcept C41008148 @default.
- W636812313 hasConcept C41681595 @default.
- W636812313 hasConcept C459310 @default.
- W636812313 hasConcept C96972482 @default.
- W636812313 hasConceptScore W636812313C119599485 @default.
- W636812313 hasConceptScore W636812313C126255220 @default.
- W636812313 hasConceptScore W636812313C127413603 @default.
- W636812313 hasConceptScore W636812313C173608175 @default.
- W636812313 hasConceptScore W636812313C206729178 @default.
- W636812313 hasConceptScore W636812313C2524010 @default.
- W636812313 hasConceptScore W636812313C2742236 @default.
- W636812313 hasConceptScore W636812313C2780165032 @default.
- W636812313 hasConceptScore W636812313C32900221 @default.
- W636812313 hasConceptScore W636812313C33923547 @default.
- W636812313 hasConceptScore W636812313C3826847 @default.
- W636812313 hasConceptScore W636812313C41008148 @default.
- W636812313 hasConceptScore W636812313C41681595 @default.
- W636812313 hasConceptScore W636812313C459310 @default.
- W636812313 hasConceptScore W636812313C96972482 @default.
- W636812313 hasLocation W6368123131 @default.
- W636812313 hasOpenAccess W636812313 @default.
- W636812313 hasPrimaryLocation W6368123131 @default.
- W636812313 hasRelatedWork W1911887900 @default.
- W636812313 hasRelatedWork W1966293529 @default.
- W636812313 hasRelatedWork W1970548269 @default.
- W636812313 hasRelatedWork W1976291838 @default.
- W636812313 hasRelatedWork W1984924324 @default.
- W636812313 hasRelatedWork W1995812446 @default.
- W636812313 hasRelatedWork W2005528765 @default.
- W636812313 hasRelatedWork W2038586186 @default.
- W636812313 hasRelatedWork W2082255081 @default.
- W636812313 hasRelatedWork W2144984987 @default.
- W636812313 hasRelatedWork W2147956657 @default.
- W636812313 hasRelatedWork W2149881812 @default.
- W636812313 hasRelatedWork W2153696524 @default.
- W636812313 hasRelatedWork W2164086179 @default.
- W636812313 hasRelatedWork W2169530632 @default.
- W636812313 hasRelatedWork W2462835504 @default.
- W636812313 hasRelatedWork W2947567876 @default.
- W636812313 hasRelatedWork W2971450774 @default.
- W636812313 hasRelatedWork W3011239373 @default.
- W636812313 hasRelatedWork W3205637741 @default.
- W636812313 isParatext "false" @default.
- W636812313 isRetracted "false" @default.
- W636812313 magId "636812313" @default.
- W636812313 workType "article" @default.