Intel

Intel ® Software Network

Intel Software TV

Tim Mattson: Floating Points Aren't Real

  • August 24, 2011
  • Timothy Mattson (Intel)
  • Download

Are you getting different numbers for calculations as you vary the number of threads? Your algorithm may be incorrect. Real numbers are nice; floating numbers are not nice. Floating numbers are not a closed set, the overflow bits need to be rounded

Expand Description Collapse Description
Are you getting different numbers for calculations as you vary the number of threads? Your algorithm may be incorrect. Real numbers are nice; floating numbers are not nice. Floating numbers are not a closed set, the overflow bits need to be rounded and that can change the results of a calculation. Tim discusses how to work with floating point numbers and resources available to help in the Intel Math Kernel Libraries.

Leave a Comment

To obtain technical support, please go to Software Support

Date Added | Popularity