fluent.com home page

   
 

Parallel Computing on NT Clusters

 

Fluent continues to advance the capabilities of its CFD software for Intel architecture workstations running Windows NT (referred to as IA/NT). With FLUENT 5 you can now utilize multiple IA/NT workstations networked together in a cluster, leveraging the compute power of multiple machines to significantly reduce the turnaround times for large CFD simulations.

Rapid advances in IA processor and memory bus speeds have contributed to improvements in parallel scaling on multiple-processor NT servers, reducing the gap in performance between traditional UNIX workstations and IA/NT systems. Fluent's experience and expertise with parallel computing on UNIX, coupled with our close partnerships with software and hardware vendors, have ensured industry-leading parallel performance on IA/NT systems.

Benchmarking results demonstrate that a network of IA/NT workstations provides a substantial boost in performance over a single-processor workstation. For example, the benchmark shown here is approximately 6.4 times faster on a cluster of four dual-processor workstations than on one processor of the same machine. Performance scaling also continues to improve on multiple-processor NT SMP machines, as shown by benchmarks on an IBM Netinfinity M10 Server. This represents a significant improvement in scaling over earlier SMP systems.

FLUENT's parallel scaling on NT clusters can dramatically reduce turnaround time for computations; scaling remains high on up to 8 processors for larger problem sets. Benchmarks performed on a cluster of Compaq Professional Workstation 6000s using Patent MPI from Genias.
#CPUs

Small Prob.
(32K cells)

Mid-size Prob.
(90K cells)

Large Prob.
(248K cells)

1
1
1
1
2
1.8
1.8
1.9
3
2.4
2.4
2.5
4
2.9
2.9
3

New multiprocessor NT machines show improved parallel scaling compared to earlier SMP systems. Benchmarks performed on a 4-processor IBM Netinfinity Server.

Development of a LINUX port for FLUENT 5.2 is well underway. Preliminary performance data from benchmarking on the Beowulf PC cluster at the Federal Energy Technology Center (FETC), in Morgantown, West Virginia, USA, has demonstrated excellent parallel scaling.

Previous Article FluentNEWS Next Article