3 Replies Latest reply on Aug 18, 2016 1:16 PM by Intel Corporation

    E7-8870 linpack

    Eric_G

      I am trying to do some benchmark on a compute node using the linpack benchmark included in mkl.

       

      The node contain 4 Intel® Xeon® Processor E7-8870 v3 and 3072 Go of RAM.

       

      I am obtaining

      40 Gflops on 1 core

      1800 Gflops for the whole node

       

      When I compare to this site http://www.nextplatform.com/2015/05/05/intel-puts-more-compute-behind-xeon-e7-big-memory/

      I am too low (it present 1 Xeon E7-8870 v3 processor at between 2073 and 2880 Gflpos).

      Can I trust this site?

       

      I saw on another site that for a Xeon E5, it had 16flops/cycle.

      If I use that number I get :

      40GFlops / (2.9 GHz*16flops/cycle) ~ 85% efficiency for 1 core.

      1800 Gflops / (2.9 GHz*16flops/cycle*72cores) ~ 55% efficiency for the node with turbo.

      1800 Gflops / (2.1 GHz*16flops/cycle*72cores) ~ 75% efficiency for the node no turbo.

      Which seems reasonable, but is it exact?

       

      Thank you.