Energy Efficient High Performance Computing Working Group SC16 Conference
 
 
 
 

Energy Efficiency Gains: Retrospective and Perspective 

ABSTRACT: 

We have already achieved major gains in energy efficiency for both the data center and the HPC equipment. For example, the PUE of the previous CSCS data center prior to 2012 was 1.8, but the current data center is about 1.25, which is a factor of ~1.5 improvement. HPC system improvements have also been very strong, as evidenced by FLOPS/Watt performance on the Green500 List. Further, the annual capital depreciation costs of HPC systems are much greater than the operational cost of electricity to keep them running. While we have seen a 2x gain from PUE and HPC system efficiency, there are 40x gains to be had from application performance improvements. We need to optimize both time and energy to solution; first, meeting requirements for time to solution and then minimizing energy to solution.

  • What are the opportunities for major gains in data center and/or HPC equipment energy efficiency for the next 5 years?
  • What capabilities do we have for optimizing both time and energy to solution?  What are the gaps?
  • What is needed moving forward?
  • Who is responsible for that forward momentum?

Moderator: 

  • Dan Reed, University of Iowa

Panelists: 

  • Thomas Schulthess, CSCS,
  • Satoshi Matsuoka, Tokyo Institute of Technology
  • John Shalf, LBNL

Title:  Eighth Annual Workshop for the Energy Efficient HPC Working Group (EE HPC WG)

Abstract:  

This annual workshop is organized by the Energy Efficient HPC Working Group (http://eehpcwg.llnl.gov/). This workshop closes the gap between facility and IT system with regards to energy efficiency analysis and improvements. For sustainable exascale computing, power and energy are a main concern, which can only be addressed by taking a holistic view combining the HPC facility, HPC system, HPC system software, and the HPC application needs. The EE HPC WG, which is a group with over 600 members from ~25 different countries, focuses mainly on cross-sectional topics traditionally not addressed by the research community.

This workshop is unique in that it provides a forum for sharing power and energy related information and research from supercomputing centers from around the world. Discussion and audience participation is encouraged. There are presentations, panels and discussions. Presenters are mostly from major governmental and academic supercomputing centers. The panels encourage discussion around more controversial topics and include panelists from supercomputing centers, academic institutions as well as the vendor community. Topics may include: future proofing your data-center, energy efficiency at exascale, open-interfaces and frameworks for power and energy management (from applications, systems, to the facility), optical interconnects and photonics, controlling power and energy through-out the software stack.


 
 
 
  Lawrence Livermore National Laboratory
7000 East Avenue • Livermore, CA 94550
  Operated by Lawrence Livermore National Security, LLC, for the
Department of Energy's National Nuclear Security Administration
     
LLNL-WEB-670983  |  Privacy & Legal Notice