The Future of Computation in Science 2012

Mini-symposium highlights industry perspectives on the future of computational science

2012 speaker photos

Frontier concepts in computational science, from cloud computing to "virtual prototyping" of new materials, were discussed by industry leaders at a mini-symposium presented Jan. 13, 2012 by Harvard's Institute for Applied Computational Science.

Five panelists bringing perspectives from hardware, software, systems, materials science and drug development discussed new approaches and applications at the event,  hosted by Cherry A. Murray, Dean of the Harvard School of Engineering and Applied Sciences, and moderated by Efthimios Kaxiras, Director of IACS and John Hasbrouck Van Vleck Professor of Pure and Applied Physics.

The 20-minute presentations by the panelists, all members of the IACS Advisory Board, preceded an extended general discussion. Presentations featured (click for details and video recordings where available) were:

  • Bijan Davari, IBM Fellow and Vice President for Next Generation Computing Systems and Technology
    VIDEO:  .FLV  |  .MP4   AUDIO: .MP3

In this presentation, the impact of some key future applications on system architecture and design will be discussed. This includes applications such as analytics, requirements for computation on large-scale uncertain data, and system-level resiliency. The applications will drive the future system architectures in unprecedented ways, as the workload optimization in many cases dictates dramatically higher throughput at reduced latency.

It used to be that to do big computations or store large datasets, you needed to build your own custom computer. Commodity workstations made it possible to build scalable clusters on the cheap, putting high-end computation within reach of scientists and engineers everywhere. These days you don't need to build anything: Just sign up for a cloud service provider with your credit card, and get to work. I'll talk about what cloud computing means for computational science and some of the things we need to work on together to get it right.

Today's parallel supercomputers have a day job, the most computationally demanding job on a modern PC, tablet or smartphone: they render the images seen by the users. Modern GPUs have outgrown their graphics heritage in many ways to emerge as the world's most successful parallel computing architecture. This is good news for computational science of because the raw computational horsepower of these chips. Today's GPUs not only render video-game frames, they also accelerate astrophysics, video transcoding, image processing, protein folding, seismic exploration, computational finance, radioastronomy, heart surgery, self-driving cars--the list goes on and on. And parallel computing isn't just a good idea, it is the only path forward for scalable computing. If your code isn't intrinsically parallel, you will not be able to tackle ever-bigger problems in the future. I will discuss some of the technological and business imperatives driving modern parallel computing, and their implications for practicing and teaching computational science.

The intent of this talk is to review the critical roles played by the digital computer revolution in accelerating understanding and applications in both sciences and engineering. "In silico paradigm" refers to the use of digital computer modeling and simulations as forerunners of early discoveries leading real and expensive experiments. We demonstrate the power of this digital approach by specific examples from chemistry and materials sciences in semiconductor technology. In addition, we will illustrate the exciting possibilities of digitally based prototyping with illustrations from physics, biotechnology, nanotechnology, weather prediction, informatics, and new alternative forms of energy.

Pharmaceutical companies have been diligent consumers of high-performance computational tools in a piecewise fashion. However, integration of these numerical methodologies across the product-development cycle has been lagging other industries.  We sketch a picture of the current and future high-performance computing design cycle.