International HPC Summer School 2016
The Future of HPC through Driving Challenges and Enabling Opportunities Keynote by Thomas Sterling
Thomas Sterling, PRACE, XSEDE, RIKEN, Compute Canada
June 2016
Slide contents
The Future of HPC through Driving Challenges and Enabling Opportunities
Projected Performance Development
Projected Performance Development
Sunway TaihuLight
Node Architecture
40 cabinets, 3.1 Pflops each
Architecture Constraints
Slide courtesy of Maciej Brodowicz, 06/16/2016
Slide courtesy of Maciej Brodowicz, 06/16/2016
Knights Landing
ANL Aurora
The Negative Impact of Global Barriers in Astrophysics Codes
Amdahl’s Law with Overhead
Head room, margins, potential innovations
Head room, margins, potential innovations
Head room, margins, potential innovations
Game Changer – Runtime System
Distinguishing Features of ParalleX/HPX
Performance Model, Full Example System
Gain with Respect to Cores per Node and Overhead; Latency of 8192 reg-ops, 64 Tasks per Core
Motivation for HPX
SpMV for parcels and memget
Wavelet Adaptive Multiresoultion
Time Required to Check if Memory Address is Local or Remote in HPX5
Time Required to Perform a Context Switch Between Lightweight Threads in HPX5
Time Required to Create a New Lightweight Thread in HPX5
Laser Interferometric Gravitational-wave Observatory (LIGO)
LIGO Chirp Filter for Signal Target
Kaplowee!!!
Discovery