Error: Failed to load processor EdGCMToc
No macro or processor named 'EdGCMToc' found

Some Notes Before You Begin

Performance: How Fast Will It Run?

The speed at which the GISS GCM runs is based primarily upon the speed of the computer's CPU. Other factors that play a role include the number of applications running at the same time, compiler optimizations, and whether or not your system is dual- or single-processor. The 64-bit CPUs in machines such as the PowerMac G5 allow the GCM to run significantly faster, since twice as many calculations are possible during one clock cycle than in the typical 32-bit systems used by most desktop computers.

The GISS GCM divides the atmosphere into a three-dimensional grid system. The version incorporated into EdGCM uses an 8 degrees X 10 degrees latitude by longitude grid system, and has nine vertical layers in the atmosphere and two ground layers. Running the climate model entails the solving of a series of complex physics equations for every cell in the grid, and a single simulated year involves many billions of calculations. Real-world performance has always been essential for the GISS GCM for research purposes, so the model was originally coded to be highly efficient. It has been further optimized to run at acceptable speeds on desktop computers without sacrificing any accuracy, but newer desktop computers will run the model the fastest.

Over the past four years, the number of simulated years per day (syears/day) for the GCM has increased more than twenty-fold on desktop Macs. An original iMac 233MHz computer could finish a 5-year simulation in about one day, while a PowerMac dual-G5, running at 2GHz, can complete two 120-year simulations in the same amount of time. As a general guideline, most simulations that would be of interest (either in the classroom or for research) need to run at least 10 simulated years. Simulations with altered forcings, such as increased greenhouse gases, must run using the predicted ocean option and require a minimum of 35 simulated years to reach equilibrium.

Table of simulated model years per day. The speed at which the GISS GCM runs on a desktop computer scales closely with CPU speed. However, changes to the microchip architecture, L2 cache levels, and compilation optimizations may also have a significant impact. We do not yet have benchmarks for Intel Macs running Mac OSX, but the dual-core design of the iMacs is somewhat faster per CPU cycle than single-core designs (based on benchmarks using Windows XP on an iMac).

Computer (CPU) CPU Speed Simulated Years / Day
iMac (G3) 233 MHz 10
eMac (G4) 800 MHz 35
PowerMac (dual-G4) 1.42 GHz 66*
PowerMac (dual-G5) 2.0 GHz 120*
Dell OptiPlex (P4) 2.8 GHz 130
iMac-Intel (Win XP) 1.8 GHz 152
PowerMac (quad-G5) 2.5 GHz 160*

*Per processor.

It is also important that you not let the computer "sleep" when the GCM is running. Sleep mode will cause the run to stop and can corrupt the files required to complete the simulation. To prevent the computer from going into sleep mode, the Energy Saver settings for your Mac (Figure 1-7) should be set to "never sleep the computer." (Setting the display to sleep is fine, and will not effect your simulations). In addition, do not check the box that allows the hard disk to sleep, as this may also damage simulation output files.

Mac energy preferences

For PCs, the appropriate power settings are set through the Control Panel (Figure 1-8). In the Control Panel, double-click on "Power Options" to bring up a dialog box to display Power Option Properties (Figure 1-9). Select the Power Schemes tab, and from the drop down menu select the "Always On" option. As with Macs, allowing the monitor to go to sleep will not affect the running of the GCM.

PC Power settings 1 PC Power settings 2

Under both Mac OS X and Windows 2000/XP/Vista you may run additional applications, such as Microsoft Word or Excel, while the GCM is running. You may even start more than one simulation at a time, although the simulations will then have to share processor time. On single-processor systems any additional applications will slow the GCM dramatically, but will not harm the simulation in any way. On dual-processor computers the impact on the speed of the run will be minimal unless you run many applications at once. Finally, you can quit the EdGCM 4D interface once a simulation is running, because the GCM runs as a separate application in the background. However, you will need to restart the EdGCM 4D interface and choose a currently running simulation if you want to pause the simulation, or to analyze the output once the run has finished.

QuickTime Video Tutorials

Note: The video tutorials for version 3.1 are being redone to reflect the significant interface changes made in this release. Please watch our web site (www.edgcm.columbia.edu) for news on the availability of the new video tutorials.

Attachments