{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

DerekChiou_EE360N_Spring2010_Lecture24

# DerekChiou_EE360N_Spring2010_Lecture24 - Lecture 24...

This preview shows pages 1–7. Sign up to view the full content.

© Derek Chiou 1 Lecture 24: Performance Evaluation and Parallel Computing (I) Prof. Derek Chiou University of Texas at Austin

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Test of size Announcements No office hours today or tomorrow Let me know if anyone is interested in office hours Wednesday after class Exam 2 is graded TAs have the exams 4/26/2010 © Derek Chiou: EE360N: Lecture 24 2
Test of size 4/26/2010 © Derek Chiou: EE360N: Lecture 24 3 Recap and Outline Recap Finish up branch prediction Started on vector machines Outline Finish up vector machines Performance measurement

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Test of size 4/26/2010 © Derek Chiou: EE360N: Lecture 24 4 Bad Ways to Compare Performance Apples and Oranges Lightly loaded machine versus production machine Sometimes hard to avoid Who gets the credit Too many variables (changed compiler, algorithm, micro- architecture) Bad/varying base-line Compare against bad case (unoptimized code, bad algorithm for machine) Changing the base-line through the paper Choice of benchmarks “embarrassing parallel” versus “pleasingly parallel” Overstates importance of some set of features No branches, no floating point, etc. Small Small footprint, fit in cache
Test of size 4/26/2010 © Derek Chiou: EE360N: Lecture 24 5 A Single Comparable Performance Number Want a single performance number for a machine Run several benchmarks How are the results aggregated together? Averaging speedups? New Machine runs A in 1 second, B in 2 seconds Reference Machine runs A in 2 seconds, B in 1 second Speedup for A is 2/1 and for B is ½ What is overall performance of New Machine? Does average speedup give a number that can be compared?

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Test of size 4/26/2010 © Derek Chiou: EE360N: Lecture 24 6 A Single Comparable Performance Number Want a single performance number for a machine Run several benchmarks How are the results aggregated together? Averaging speedups?
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### Page1 / 27

DerekChiou_EE360N_Spring2010_Lecture24 - Lecture 24...

This preview shows document pages 1 - 7. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online