Session: Practical Supercomputing
  Title:
  Chair: Mike Calise (ClearSpeed Technology Inc.)
  Time: Thursday, November 20, 12:00PM - 1:00PM
  Rm #: 42-43
  Speaker(s)/Author(s):  
  Mike Calise (ClearSpeed Technology, Inc.), Wu-chun Feng (Los Alamos National Laboratory), Erik DeBenedictis (Sandia National Laboratories), Srinidhi Varadarajan (Virginia Polytechnic Institute), James Phillips (University of Illinois), David Rich (Advanced Micro Devices), Ron Neyland (RLX), Paul Terry (Octigabay Systems), Padmanabhan Iyer (Linux Networx)
   
  Description:
  Traditionally, supercomputing has had to focus on performance at the expense of other issues such as power consumption, space requirements and cost of ownership. Present supercomputers that provide computational resource measured in the tens of Teraflops/s (10,000 processors) are presently quoted in Mega Watts and in football field sized dimensions. Such an approach has limited the use of these systems to a few specialized organizations.



However, the application areas of supercomputing are now growing rapidly. Computational simulation is playing an increasingly important role in basic science, both in accelerating and substituting for physical experimentation and also for prediction where physical experimentation is difficult, costly or impracticable. Digital content creation is growing significantly with the realization of HDTV. Applications such as computational biology and nanotechnology researching first principles molecular dynamics are characterized by their need for almost unlimited high performance computing needs.



To allow such applications to become widely available a multi Teraflop/s computing machine must be within the power and physical size limits of a moderate machine room installation, be easily programmable and offer a price performance ratio in line with commodity PC levels. For such practical supercomputing, a radical re-think is required.



Is this possible? Can we really achieve the goal of affordable, low power TeraFlops? Is there a market? Who will drive it? Can HPC become pervasive and in the hands of the masses?



The objective of this session is to bring together parties interested in discussing the requirements and challenges for a more practical approach to supercomputing. The floor will then be opened for an informal discussion on this topic.



  Link: --