Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: real applicaEons –  e.g., memory bandwidth CSEP 524: Parallel ComputaEon Winter 2013: Chamberlain 13 (Currently #1 on the Top 500) •  •  •  •  •  •  Titan compute nodes: 18,688 processors: 16- core AMD/node = 299,008 cores GPUs: 18,688 NVIDIA Tesla K20s memory: 32 + 6 GB/node = 710 TB total peak speed: 20+ petaflops floorspace: 4,352 square feet For more informaEon: hbp://www.olcf.ornl.gov/Etan/ CSEP 524: Parallel ComputaEon Winter 2013: Chamberlain 14 “Glad I’m not an HPC Programmer!” A Possible Reac3on: “This is all well and good for HPC users, but I’m a mainstream desktop programmer, so this is all academic for me.” The Unfortunate Reality: –  Performance- minded mainstream programmers will increasingly deal with parallelism –  And, as chips become more complex, locality too CSEP 524: Parallel ComputaEon Winter 2013: Chamberlain 15 My Mac’s Processor: an Intel Core i7 CSEP 524: Parallel ComputaEon Winter 2013: Chamberlain Source: hbp://www.legitreviews.com/arEcle/1484/1/ 16 Some Hardware Terminology processor core (or simply “core”): the unit of a computer that has a PC, executes instrucEons, etc. (compute) node: a group of cores and memories that must go over a network to communicate with any others network: the wires and chips that permit nodes to communicate with one another memory memory c c c socket node CSEP 524: Parallel ComputaEon c network Winter 2013: Chamberlain c c c socket node c 17 More HW Terms: Shared vs. Distributed Memory shared memory: A system in which memory can be accessed via simpl...
View Full Document

This document was uploaded on 04/04/2014.

Ask a homework question - tutors are online