Program: SC2002 Gordon Bell Awards

SC2002 Gordon Bell Awards

The Gordon Bell Prizes
The Gordon Bell Prizes are awarded each year at the annual SC conference to recognize outstanding achievement in high performance computing. The $5,000 prize is donated by Gordon Bell, a pioneer in computer architecture, parallel processing, and high performance computing. The goal of the award is to stimulate future advances of parallel computing applications by identifying major accomplishments and tracking progress over time. Prizes are awarded in three categories:

1. Special Accomplishment: The prize in the special accomplishment category may be given to the entry that utilizes innovative techniques to demonstrate the most dramatic gain in sustained performance for an important class of real-world application. Such techniques may include advances in mathematical algorithms, data structures, or hardware or software implementations. The measured performance need not surpass or meet the level attained for the entry that wins the Peak Performance prize but rather represent the most important achievement in high performance computing considered by the Bell Prize Committee.

2. Peak Performance: The prize in the peak performance category is given to the entry demonstrating the highest performance achieved in terms of operations per second on a real-world application program. Recent winners have demonstrated performance in the range of a teraflops.

3. Price/Performance: The prize in the price/performance category is given to the entry demonstrating the best cost to performance ratio as measured in megaflops per dollar on a real-world application.

Depending on the entries received, in some years no prize will be awarded in a given category. A listing of previous Gordon Bell Awards is available here.

Entering the Contest
Entries in competition for the Gordon Bell Prize should be submitted as regular papers to SC2002. Such a paper must be specifically identified as a Gordon Bell Prize entry and must specify which single prize category it is to be considered for. Finalists will be selected by the five-person Bell Prize Committee and notified at the same time that paper acceptances are announced. Finalists will then have an opportunity to improve their results and resubmitted them at the same time as technical papers. The winners of the 2002 Gordon Bell Awards will be selected by the Bell Prize Committee based on these submitted revised results. Winners will be announced at SC2002.

Advice to Submitters
Those intending to enter the Gordon Bell Prize competition should familiarize themselves with the high standards set by previous winners, which reflect some of the best work being done in high performance computing by the world community. A summary is given below. One can find additional details on previous winning submissions by consulting papers published on earlier competitions. (e.g. Alan H. Karp, Ewing Lusk, David H. Bailey, “1997 Gordon Bell Prize Winners,” IEEE Computer, vol. 31, no. 1, January, 1998, pp. 86-92) included in the Proceedings of the SC98 through SC01 Conferences. Critical to the success of any submission for all categories is a detailed and precise explanation describing how performance was measured, and, in the case of the price-performance category, how costs were estimated. In this latter case, it is important that there be no hidden costs not accounted for. For all three categories, it is important that the computation represent a genuine real-world application, one used to perform real end user work, not simply a demonstration or benchmark code. Further, these applications must reflect significant challenges to effective high end computing.

For example, embarrassingly parallel calculations requiring little or no global data communications have generally not been selected as finalists in previous years. Also important is that the algorithms be efficient, that is, there be minimal redundant computations and that the number of operations required to perform a given computation be near minimum. Therefore, for all entries, the algorithm must be described in sufficient detail such that the committee can accurately assess the nature of the computational work being performed. Professional practices for citing performance must be observed – see for example the essay “Twelve Ways to Fool the Masses.” For the Special Accomplishments category, where dramatic performance gain is sought for a specific application class, it is imperative that the base-level performance of the conventional problem to which the new results are being compared be sufficiently documented and references sited so that the committee may fairly assess the achievement being submitted.

Specific questions can be addressed to: bellawards@sc-conference.org

GORDON BELL AWARDS CHAIR
Thomas Sterling
California Institute of Technology and the Jet Propulsion Laboratory