ANSYS announced last week that an Italian research team successfully conducted a
1-billion cell simulation to investigate the aerodynamics of an America’s Cup yacht, reaching
a milestone in the simulation world. To give context, a typical ANSYS Multiphysics
benchmark for a sophisticated computing cluster runs to about 6 million cells, uses 2
Gigabytes of memory and takes about 5 to 10 minutes to iterate to a solution, depending
upon the number of processors and how they are all connected to one another and to storage

But the “typical” benchmark is just that – it’s not intended to push the envelope of what is
possible. As simulation becomes increasingly realistic, as analysts try to solve their
real-world problems rather than a simplified subset, simulations are becoming ever-larger and
resolving this scale of problem will become common-place. Proving that the simulation
industry’s “stretch goal” of 1 billion cells can be solved is a massive step forward.

This 1 billion cell simulation was performed on CILEA’s (an Italian inter-university consortium
for information and communication technologies, named Consorzio Interuniversitario
Lombardo per L’Elaborazione Automatica) Lagrange supercomputer, ranked #135 on the list
of the 500 most powerful supercomputers in the world. Lagrange is an HP Cluster Platform
3000BL running Linux, with 208 HP ProLiant BL460c server blades and Intel Xeon 3.166 GHz
quad-core central processing units. Total peak performance of the system approached 22
teraflops (22 thousand billion floating point operations) per second; the system has recently
been ranked number 135 on TOP500’s list of supercomputers.

If you remember past races, you’ll know that America’s Cup yachts often involve weird and
wonderful underwater structures. This leads to, as ANSYS says, “some of the most complex
physics effects possible, with hydrodynamic and aerodynamic fluid flow and stiffness among
the structural physics involved.” The test-case simulation focused on the aerodynamic impact
of wind on the racing yacht sailing downwind, with a particular emphasis on the mainsail and
an asymmetrical spinnaker.

The test case was completed in 170 hours, which seems like a lot. But ANSYS
characterizes this time frame as “commercially viable”, given that the value of the solution is
mission critical and expensive (if possible at all) to arrive at in any other way. Typically, in
yachting, a scale model is built and tested in a wind tunnel; ANSYS reports that the
simulation results showed “good agreement” to the wind tunnel results. The question, of
course, is how much it cost and how long it took to build the scale model and test it in the
wind tunnel – and how that relates to the cost and time of creating and executing the
simulation.

ANSYS has some pretty cool images available at
http://www.ansys.com/special/news-images/2008/billion-cell-11-17-imagesheet-08.htm.

While this 1-billion cell case was clearly a test to see if this was possible at all, it’s not a
stretch to say that larger and larger analyses are run every day all over the world. Companies
are investing heavily in the compute infrastructure to support their analysts, with the view that
realistic analyses will improve product reliability, reduce warranty exposure and minimize
risk. Not a bad investment.