NERSCPowering Scientific Discovery Since 1974

Xeon Phi Users Group: Performance Tuning and Functional Debugging for Xeon Phi

November 19, 2014

 Intel Xeon Phi Users Group SC14 BoF Session 

LOCATION: At SC14 in New Orleans, LA, Nov. 19, 2014. 

SC14 Web Linkhttp://sc14.supercomputing.org/schedule/event_detail?evid=bof176

EVENT TYPE: Birds of a Feather

TIME: 5:30PM - 7:00PM Central Time

SESSION LEADER(S)Richard Gerber, Kent Milfeld, Chris Gottbrath, Chris J. Newburn

ROOM:286-87

Abstract:

This BOF will build community among those developing HPC applications for systems incorporating the Intel Xeon Phi many-core processor. Taking advantage of the processor’s full capabilities requires tuning and optimizing using programming techniques and tools targeted at a combination of CPUs and the Xeon Phi Coprocessor. Threading, vectorization, memory contiguity and alignment, and data locality may be important for performance. This BOF will combine a few brief presentations sharing insights and best practices with a moderated discussion among all those in attendance. It will close with an invitation to an ongoing discussion through the Intel Xeon Phi Users Group (IXPUG).

Long Description

This BOF will serve as a gathering point for those developing HPC applications for systems incorporating the Intel Xeon Phi many-core processor. Participants will share insights that come from performance tuning and debugging on those systems. Taking advantage of the processor’s full capabilities requires tuning and optimizing using programming techniques and tools targeted at a combination of CPUs and the Xeon Phi Coprocessor. Even applications that are already multi-threaded may need to be reworked to expose additional concurrency. Developers also need to pay attention to threading, vectorization, memory contiguity and alignment, and data locality to get the performance they are looking for. BOF participants will share insights and experiences with performance tuning and debugging on these systems.

This BOF will start with brief presentations from academia, national labs and industry representatives sharing their insight about tuning and debugging on the Xeon Phi. It will then proceed to a moderated discussion of the topics raised among those in attendance.

Participants will be introduced to an independent international user group for those developing HPC applications for the Xeon Phi. The Intel Xeon Phi Users Group (IXPUG), based at TACC, is expanding world wide to include all HPC sites that incorporate the Phi, whose growing importance in HPC is evidenced by the two Phi-based systems in the top 10 of the June 2014 Top 500. NERSC and LANL recently announced large procurements of next-generation Phi systems. TACC, NERSC, and LANL will lead a discussion with attendees to plan future meetings and activities of the users group.

In addition to user group planning, the work product output of this BOF will be a summary of the talks, with detailed notes on points brought up during the BOF, and the results of a five-question survey taken during the BOF to gather feedback and user experiences beyond those which were covered in the BOF.

The BOF organizers, Richard Gerber of NERSC, Kent Milfeld of TACC, Chris Gottbrath of Rogue Wave, and Chris J. Newburn of Intel will kick off the BOF, moderate presentations and discussion, and frame follow up activities.

Agenda  

Time (CST)TopicSpeaker
5:30  Welcome and Overview Kent Milfeld, TACC; Chris Gottbrath, Rouge Wave; CJ Newburn, Intel; Richard Gerber, NERSC
5:40 Performance Optimization of the Amber Molecular Dynamics code

Manoj Nambiar, Principal Scientist, TCS Innovation Labs, Mumbai, India

5:45

Scaling HPC-BLAST

Glenn Brook, University of Tennessee, Knoxville

5:50 Hierarchical Equations of Motion:  OpenCL on the Xeon Phi  Matthias Noack, Zuse Institute Berlin (ZIB), Germany
5:55 Large-Scale Hydrodynamic Brownian Simulations Edmond Chow, Georgia Institute of Technology
6:00 Refactoring the HBM 3D Ocean Model Jacob Weismann Poulsen, DMI, Denmark
6:05 MAGMA MIC: HPC Linear Algebra for Intel Xeon Phi Piotr Luszczek, University of Tennessee, Knoxville
6:10 Open / Panel Discussion  All attendees and Lightening Talk Participants. Moderated by CJ Newburn, Intel and Chris Gottbrath, Rouge Wave
6:45 IXPUG Discussion Led by Richard Gerber, NERSC
7:00 Conclude  

 

Session Leader Details:

Richard Gerber (Primary Session Leader) - National Energy Research Scientific Computing Center, Berkeley Lab

Kent Milfeld (Secondary Session Leader) - TACC, University of Texas at Austin

Chris Gottbrath (Secondary Session Leader) - Rogue Wave Software

Chris J. Newburn (Secondary Session Leader) - Intel Corporation