Significant performance improvement of symmetric eigensolvers and SVD in Intel MKL 11.2

 

Intel MKL 11.2 contains a number of optimizations for Symmetric Eigensolvers and SVD. These mostly related to large matrices N>4000, 6000, and on but speedups are significant comparing to the previous MKL 11.1.  SVD brings up to 6 times (or even higher on large thread counts and matrix sizes), similarly for eigensolvers, several times could be observed.

List of related optimizations present in MKL 11.2 are:

  • Improved performance of ?(SY/HE)(EV/EVR/EVD) when eigenvectors are not needed
  • Improved performance of ?(SY/HE)(EV/EVD) when eigenvectors are needed
  • Improved performance of ?(SY/HE)RDB               
  • Added Automatic Offload for ?SYRDB on Intel® Many Integrated Core Architecture (Intel® MIC Architecture), which speeds up DSY(EV/EVD) when eigenvectors are not needed
  • Improved performance of (S/D)GE(SVD/SDD) when M>=N and singular vectors are not needed

Below performance charts showcases the performance improvements for DGESVD and DSYEV routines on Intel® Xeon® E5-2600 processors.

DGESVD MKL 11.2 performance improvement

DSYED MKL 11.2 performance improvement

We would like to get your feedback.  Please use the comment box below to provide us feedback or submit it in http://premier.intel.com

 

For more complete information about compiler optimizations, see our Optimization Notice.