This Saturday, 21 April 2012, was the New England Numerical Analysis seminar in Amherst.
Of course I was there. The whole thing was really interesting but I will only comment on what caught my attention. The first presentation was about simulations of supernovas and had some really cool videos. The speaker was Robert Fisher from UMass Dartmouth. The scale of those simulations simply blew my mind! Another potential application for GPU computing. 😉
Here’s the page to his group.
Kirk Jordan of IBM Research gave a talk on ExaScale computing. What really struck me was the statement that as cores keep on becoming more and more on the same chip, soon we will have errors in our programs. Numerical errors that the software will not be able to catch and, while we wait for a result for days/months/weeks/years, those come along with the errors and we just have wasted our time. Scary stuff!!
Then Lorena Barba of BU talked about fast multipole methods and how they make a difference in today’s scientific computing. Professor Barba is behind the ExaFMM code, a program which runs on NVidia gpus using CUDA and uses them to accelerate the Fast Multipole method. It was great that I had the chance to talk to her, since I will be working on that project during the summer along with one of my professors, Professor Hans Johnston.
Lastly, I really liked the presentation that Professor George Karniadakis gave and which touched several subjects, the highlight in my opinion being the simulation of a whole human, from the DNA level to the blood, to the organs and at last to the body. So many different scales on which you must do simulation and then connect the different scales together. The project is of course huge but that’s why it’s interesting! Imagine all the data needed!
I will leave you with something that was mentioned by both professor Barba and professor Johnston, a quote by professor Trefethen of Oxford university. Then again, this is a blog, and I can post all of them. They come from an essay called “Predictions for scientific computing fifty years from now”
We may not be here.
We’l l talk to computers more often than type to them, and they’ll respond with pictures more often than numbers.
Numerical computing will be adaptive, iterative, exploratory, intel ligent and the computational power will be beyond your wildest dreams.
Determinism in numerical computing will be gone
The importance of floating point arithmetic will be undiminished
Linear systems of equations will be solved in flops
Multipole methods and their descendants will be ubiquitous.
. Breakthroughs will have occurred in matrix preconditioners, spectral methods, and time stepping for partial dierential equations.
The dream of seamless interoperability will have been achieved.
The problem of massively parallel computing will have been blown open by ideas related to the human brain.
Our methods of programming will have been blown open by ideas related to genomes and natural selection
And for the last one,
If we start thinking now, maybe we can cook up a good name for our field!
I recommend reading the whole essay which is here.