Ocean predictability studies in a parallel computing environment. Final report
The goal of the SDSC effort described here is to evaluate the performance potential of the Oberhuber isopycnal (OPYC) ocean global circulation model on the 64-node iPSC/860 parallel computer at SDSC and its near term successor, the Intel Paragon, relative to that of a single vector processor of a CRAY Y-MP and its near term successor, the CRAY C90. This effort is in support of a larger joint project with researchers at Scripps Institution of Oceanography to study the properties of long (10 to 100-year scale), computationally intensive integrations of ocean global circulation models. Generally, performance of the OPYC model on the iPSC/860 has proved quite disappointing, with a simplified version of the entire model running at approximately 1.4 Mflops on a single i860 node in its original form and after extensive optimization, including coding in assembler, at 3.2 Mflops. The author estimates overall performance to be limited to 75 Mflops or less on the 64-node machine, as compared to 180 Mflops for a single CRAY Y-MP processor and 500 Mflops for a single CRAY C90 processor. Similarly, he believes an implementation on an Intel Paragon, even with several hundred nodes, will not be competitive with a single processor C90 implementation. Additionally, the Hamburg Large Scale Geostrophic (LSG) model has been implemented on high-performance workstations and evaluated for its computational potential as an alternative to OPYC for the long term integrations.
- Research Organization:
- San Diego Supercomputer Center, CA (United States)
- Sponsoring Organization:
- USDOE Office of Energy Research, Washington, DC (United States)
- DOE Contract Number:
- FG03-91ER61217
- OSTI ID:
- 505314
- Report Number(s):
- GA--A21159; ON: DE97007847
- Country of Publication:
- United States
- Language:
- English
Similar Records
Lattice gauge theory on the Intel parallel scientific computer
Fast parallel algorithms for short-range molecular dynamics
Neural network simulations on massively parallel computers: Applications in chemical physics
Conference
·
Wed Aug 01 00:00:00 EDT 1990
· AIP Conference Proceedings (American Institute of Physics); (USA)
·
OSTI ID:5910225
Fast parallel algorithms for short-range molecular dynamics
Technical Report
·
Sat May 01 00:00:00 EDT 1993
·
OSTI ID:10176421
Neural network simulations on massively parallel computers: Applications in chemical physics
Conference
·
Thu Dec 31 23:00:00 EST 1992
·
OSTI ID:6649368