Home >
CSM 2005 Highlight Archive
Benchmarks show Cray X1 suitable for large-scale molecular dynamics simulations --
a first step in understanding lubrication
 There is an enormous variety of problems
involving MD that require large amounts of computation, both major extensions of
existing studies and exploration in new fields that have yet to be tackled because
of a lack of adequate high-performance computing capability. Recent studies using
vectorization of Rapaport’s scalable domain-decomposed MD code show essentially
perfect scaling up to the full size of the Cray X1 (512 MSP nodes at the time
of the runs) at NCCS -- demonstrating the capability of the Cray X1 to deal with
problems of this type. more
New algorithms provide definitive molecular electronic structure AND record processing speed
The Full-configuration interaction (FCI) method developed by Z. Gan and R.J. Harrison
exactly solves the N-electron problem in a given 1-electron basis. This solution provides
a critical benchmark for understanding and calibrating many-body calculations. The researchers have
developed several new parallel-vector algorithms, each optimal for certain parameter values.
The optimal algorithms for Cray architectures differ sharply from those for IBM. The new code
already enables computations 40+ times larger and 100+ times faster than previous work and is
achieving 62% of peak on the ORNL Cray X1 (3.4 TFLOP/s sustained on 432 MSP).
ORNL at SC|05
|
