chapter1-AppendixA-m4-ziavras

Culler d patterson ucb i i s ziavras example

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: enchmark suite? • Decide if mean a good predictor by characterizing variability of distribution using standard deviation • Like geometric mean, geometric standard deviation is multiplicative rather than arithmetic • Can simply take the natural logarithm of SPECRatios, compute the standard mean and standard deviation, and then take the exponent to convert back: ⎛1 n GeometricM ean = exp ⎜ × ∑ ln (SPECRatio ⎝ n i =1 GeometricS tDev = exp (StDev (ln (SPECRatio Some material is adapted from D. Culler & D. Patterson (UCB) i )⎞ ⎟ i ))) ⎠ S. Ziavras Example Standard Deviation (1/2) • GM and multiplicative StDev of SPECfp2000 for Itanium 2 Periods missing in plot (class viewgraphs) 14000 SPECfpRatio 12000 10000 GM = 2712 . GSTEV = 1.98 8000 6000 5362 . 4000 2712 . 2000 1372 . apsi sixt rack fma3d lucas ammp fac erec equake art galgel mesa applu mgrid s wim wupwise 0 Single standard deviation range: [GM / STDEV, GM X STDEV] Some material is adapted from D. Culler & D. Patterson (UCB) S. Ziavras Example Example Standard Deviation (2/2) • GM and multiplicative StDev of SPECfp2000 for AMD Athlon Periods missing in plot 14000 SPECfpRatio 12000 10000 GM = 2086 . GSTEV = 1.40 8000 6000 4000 2911 2086 1494 2000 apsi fma3d lucas ammp facerec equake art sixtrack Some material is adapted from D. Culler & D. Patterson (UCB) galgel mesa applu mgrid swim wupwise 0 S. Ziavras Comments on Itanium 2 and Athlon • Standard deviation of 1.98 for Itanium 2 is much higher-- vs. 1.40--so results will differ more widely from the mean, and therefore are likely less predictable • Falling within one standard deviation: st – 10 of 14 benchmarks (71%) for Itanium 2 – 11 of 14 benchmarks (78%) for Athlon Some material is adapted from D. Culler & D. Patterson (UCB) S. Ziavras Summarizing Performance Summarizing Performance • Arithmetic mean tracks execution time: mean execution time: Σ(Ti)/n or Σ(Wi*Ti) • Harmonic mean of rates tracks execution time: m/Σ(1/Ri) or n/Σ(Wi/Ri) • Normalized execution time used for scaling performance (e performance (e.g., X times faster than SPARC) times faster than SPARC) – But do not take the arithmetic mean of normalized execution time, use the geometric mean: ( Π Tj / Nj )1/n Some material is adapted from D. Culler & D. Patterson (UCB) S. Ziavras...
View Full Document

Ask a homework question - tutors are online