introduction to the xc30
Download
Skip this Video
Download Presentation
Introduction to the XC30

Loading in 2 Seconds...

play fullscreen
1 / 10

Introduction to the XC30 - PowerPoint PPT Presentation


  • 106 Views
  • Uploaded on

Introduction to the XC30. For Users of Cray’s XT5 and XK7 Aaron Vose. XT5/XK7 to XC30 Changes Overview. The hardware is similar, with changes: CPU: Move to from AMD Istanbul/ Interlagos to I ntel SNB/IVB AMD’s CMT vs. Intel’s HyperThreading Interconnect :

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Introduction to the XC30' - peyton


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
introduction to the xc30

Introduction to the XC30

For Users of Cray’s

XT5 and XK7

Aaron Vose

slide2
XT5/XK7 to XC30 Changes Overview
  • The hardware is similar, with changes:
    • CPU:
      • Move to from AMD Istanbul/Interlagos to Intel SNB/IVB
      • AMD’s CMT vs. Intel’s HyperThreading
    • Interconnect:
      • XT5’s SeaStar or XE/XK7’s Gemini -> XC30’s Ares
  • The software and environment is very similar as well
slide6
Intel’s HyperThreading vs. AMD’s Modules
  • AMD CPU module contains 2 integer cores sharing an FPU:
    • Allocate cores sharing FPU by default with 16 threads/CPU
    • Can get a dedicated FPU per thread by running 8 per CPU
  • Intel CPU can run 2 threads per core with “HyperThreading”:
    • aprun -j 2 -> Two ranks per core (“DualStream”)
    • aprun -j 1 -> One rank per core (“SingleStream”)This is the default
slide7
XC30’s Ares Interconnect
  • Switch from 3D Torus to Dragonfly topology
    • Adaptive routing avoids network hotspots
    • Node placement not as important
    • Greatly increased global bandwidth
  • Fill me in
slide8
Software Changes: ACML -> MKL
  • AMD Core Math Library is gone; replaced with:
  • Intel’s Math Kernel Library
    • Update code:
      • call vrda_exp(VL,RF(1,1),RF(1,1)) (ACML)
      • call vdexp(VL,RF(1,1),RF(1,1)) (MKL)
    • To use MKL:module load intel
    • Link with: -L$(MKLROOT)/lib/intel64/ -lmkl_intel_lp64 -lmkl_sequential -lmkl_core
slide9
Software Changes: Intel Compiler
  • Intel Compiler: module swap PrgEnv-* PrgEnv-intel
    • ftn/cc/CC commands wrap ifort/icc/icpc
  • Useful Flags:
    • -openmp(Enables OpenMP)
    • -xAVX(Enables AVX)
    • -mkl(Enables MKL)
slide10
XT5/XK7 to XC30 Changes: Summary
  • Hardware:
    • Move to Intel SandyBridge / IvyBridge
    • Intel HyperThreading
    • Interconnect: Ares -- Dragonfly Topology
  • Software and environment:
    • Intel MKL
    • Intel Compiler
ad