Architecture compatability for the Deep Learning Performance Optimization:

Architecture compatability for the Deep Learning Performance Optimization:

As per the link-https://cug.org/proceedings/cug2017_proceedings/includes/files/tut110s2-file3.pdf, improved performance metrics are identified on intel architecture which is of "2 sockets broadwell -22 core" category.

Do we need to look out for same configuration "2 sockets broadwell -22 core" to obtain the same result?

Can I use the following PROCESSOR for my own evaluation? - "Intel(R) Xeon(R) CPU E5-2680 v2"

Can you please let us know the Intel processors (Processor number) that are capable of producing the same result?

Zone: 

2 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

Hi,

Intel(R) Xeon(R) CPU E5-2680 v2 is a 10 core part so performance will be lower than the 2 x 22 core (dual socket) setup mentioned in the slide deck. As one can expect, the same result cannot be obtained by a lower core count part of the same architecture.

Please note that MKL is optimized for all BDW and SKL variants so performance should be significantly improved with the MKL integrated version of TF as compared to the non-MKL version.

p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; font: 11.0px Helvetica; color: #1f497d}
span.s1 {font-kerning: none}

Leave a Comment

Please sign in to add a comment. Not a member? Join today