Home > climate models > PCCM2.1 Performance Implementations

PCCM2.1 Performance Implementations

2012 March 6


PCCM2.1 Performance Implementations of PCCM2 are now available on the Intel Paragon, the IBM SP2, the CRAY T3D and the CRAY C90. The performance of of the code on these platforms gives an indication of relative machine performance and indicates the usefullness of the MPP platforms for climate research. Two horizontal resolutions are compared The performance is given in the number of seconds it takes to simulate one day on the computer. Two resolutions are shown. The T42 resolution corresponds to a 2.8 degree grid while T170 is a 0.7 degree grid. The higher resolution represents the what might be required for regional scale modeling of climate impacts.

http://www.csm.ornl.gov/chammp/highlights.html

I don’t often use ‘OMFG’ in conversation. We have all heard of Moore’s Law; but when you see it action, OMFG seems appropriate.

My 2-3 year old, off-the-shelf Dell Studio 1747 laptop matches the performance of NCAR’s Cray T3D parallel supercomputer in service from 1994-1999. The laptop is the purple addition on the left hand side to the PCCM2 Performance chart.

I was also curious about the performance hit that might be caused by running in a VirtualBox VM. So I repartioned, installed Ubuntu to the hard disk, and reran the model directly on the host os. There was only the smallest of improvements: from 34minutes on the VM to 33 minutes on the hardware for a 10 day run. Less than a minute difference.

More info on NCARs Cray T3D supercomputer can be found here:
http://www.cisl.ucar.edu/computers/gallery/cray/t3d.jsp

===========

Thinking about OpenCL / CUDA in a climate modeling paradigm. Feel free to toss pointers my way.

Advertisements
  1. ds
  2. 2012 March 8 at 2:47 pm
  1. No trackbacks yet.
Comments are closed.