Programming Methodologies

Published on Sep 19, 20161191 Views

IHPCSS 2016 - Ljubljana

Programming Methodologies00:00

FLOPS we need: Climate change analysis00:26

Qualitative Improvement of Simulation with Higher Resolution (2011)02:03

Exascale combustion simulations03:31

Modha Group at IBM Almaden04:18

Waiting for Moore’s Law to save your serial code start getting bleak in 200405:33

Moore’s Law is not at all dead…06:58

At end of day, we keep using all those new transistors.08:02

That Power and Clock Inflection Point in 2004… didn’t get better.08:18

Not a new problem, just a new scale…09:54

How to get same number of transistors to give us more performance without cranking up power?10:30

And how to get more performance from more transistors with the same power.10:59

Prototypical Application: Serial Weather Model12:07

First parallel Weather Modeling algorithm: Richardson in 191712:38

Weather Model: Shared Memory (OpenMP)13:37

Weather Model: Accelerator (OpenACC)14:27

Directives: an elegant and common approach.16:46

Weather Model: Distributed Memory (MPI)18:00

The pieces fit like this…19:38

Other Paradigms?21:11

Top 10 Systems as of June 201626:31

Today28:50

Projected Performance Development29:22

Trends with ends.29:47

Two Additional Boosts to Improve Flops/Watt and Reach Exascale Target30:57

End of Moore’s Law Will Lead to New Architectures33:36

It would only be the 6th paradigm.35:32

We can do better. We have a role model.36:08

Again…36:55

The Future and where you fit.37:42

Appendix42:15