March 14, 2008

HPC on the Fast Track

Michael Feldman

Formula One racing seems to be on an HPC tear lately. Last week, we covered the purchase of an Appro system for Renault's F1 Team. This week, we look at how Red Bull Racing is using Platform LSF to get the most out of their three cluster systems.

Over the past five years, high performance computing has established itself as a mainstream technology for Formula One (F1) race car design. Because of the nature of the sport, Formula One requires extremely sophisticated engineering. Considered the elite form of auto racing (sorry NASCAR), these cars reach speeds in excess of 200 mph and run on the most challenging racing circuits in the world. Today, most of the top tier F1 teams have turned to HPC to accelerate race car development, especially aerodynamic design.

Last Friday, Red Bull Racing became the latest F1 team to announce an expansion of their HPC commitment. Red Bull Technology, the designer and manufacturer of Red Bull Racing’s Formula One cars, announced they would be adding Platform LSF (Platform Computing’s workload manager). The team will use LSF to schedule CFD simulation jobs across the group’s three IBM compute clusters.

The systems used by the Red Bull Racing team consist of two smaller clusters, with about 250 cores apiece, and a larger 1,024-core machine — all based on AMD Opterons. It wasn’t until they installed the large cluster last year that they started looking seriously at the Platform LSF product. At that scale, it become necessary to do a better job at managing all the CFD simulations. Manual submission of the jobs became impractical with the larger machine and with multiple systems. By adding LSF, the technology team is able to merge the clusters into one virtual system, enabling users in multiple departments to share computing resources.

I got the opportunity to speak with Steve Nevey, business development manager for Red Bull Technology, who gave me a sense of how critical HPC has become to Formula One teams, and to Red Bull’s in particular. Nevey’s own career has paralleled the rise of computing in Formula One. Originally a design engineer in the shipbuilding industry, he got into Formula One racing about 20 years ago as a CAD specialist with the Arrows team, which had an active F1 program from 1977 to 2002. In 1996, he joined Jackie Stewart’s new Formula One team (Stewart Grand Prix), the precursor of the current Red Bull team.

The Stewart Grand Prix team was set up at the invitation of the Ford Motor Company, and in 1999 the team was bought by Ford to become Jaguar Racing. For the next five years, they raced under the Jaguar brand. In 2005, Ford put the team up for sale, at which point it was acquired by Red Bull, one of the sponsors of Jaguar Racing. That same year the team competed under the Red Bull Racing name for the first time. Also in 2005, Red Bull bought the Minardi team and renamed it Scuderia Toro Rosso. The two Red Bull teams have been racing ever since.

When Nevey was originally hired as the IT manager of Stewart Grand Prix in 1996, he was just a CAD engineer. “But I was the first person to walk through the door who knew anything about computers,” he told me. “So they made me the IT manager.”

He did that for about 18 months until they hired a “proper IT manager,” at which point he was able to concentrate on engineering again. About five years ago, Nevey transitioned into more of a commercial role. Now, as the business development manager of the Red Bull Team, he’s responsible for identifying and managing partnerships with their various technical partners and suppliers, which includes companies such as ANSYS (Fluent), Siemens, MSC.Software, and now, Platform Computing.

When Nevey started with Stewart Grand Prix, they had 15 design engineers, which were doing mostly CAD work at individual workstations. Five or six years ago, they introduced computational fluid dynamics (CFD) into the engineering design workflow. At that time, they were just following the trend of other Formula One teams, like McClaren and Ferrari, who had started playing around with vehicle simulations.

“It was something we didn’t fully understand or understand the value of,” said Nevey. At the time, the team’s aerodynamic engineers were telling management that the CFD simulations took too long to be really useful — they had less than a ten-node system at the time — and they couldn’t validate the results. It soon became apparent that they had no justification to use HPC for race car development, so they shut down the system.

Undeterred, they developed a business plan to show how the use of HPC could be cost-effective for the program and raise the bottom line. The plan integrated the CFD simulation work into the overall development process, maximizing both vehicle design work and wind tunnel testing. Today, Nevey says the Red Bull Technology team is up to about 150 engineers. “CFD is now absolutely vital to what we’re doing,” he said. “If we didn’t have it, it would leave a big gap.”

In Formula One culture, the conventional wisdom is that the CFD simulations don’t replace the wind tunnel; it just allows a lot more design iterations to take place before scaled-down (60 percent) components get built and sent to the tunnel for validation. After the tunnel, full-sized parts are constructed, which are then installed on the vehicle for final testing. Wind tunnels tend to be in constant use, so the more design work you can do inside the computer, the better.

CFD is used to design body components in such a way as to balance the aerodynamics of downforce and drag. Downforce is created by the wings and other aerodynamic components of the car to push it down onto the track (opposite of what occurs on airplane wings). At high speeds, this means the weight of the car is up to four times heavier than its weight while at rest. More downforce puts greater load into the tires for better grip, which allows for better cornering. But a wing design that maximizes downforce, also raises drag, which slows the vehicle down on the straight sections of the circuit. The CFD engineers are constantly balancing the compromise between the two.

The calculation for specific body components has to take into account individual racing circuits. Unlike NASCAR, which is typically run on oval tracks, F1 circuits are quite variable in shape and features. For example, the Monaco Grand Prix is known for being an extreme high downforce circuit, with lots of tight corners. So the engineers will be looking at big wings to push the car onto the track, with less emphasis on drag. Toward the end of the season, the lowest downforce circuit, the Autodromo Nazionale Monza in Italy, will require body components that produce a much lower aerodynamic profile for those long straight sections.

Racing conditions such as weather and the amount of rubber on the track (from previous races) can also effect race time design changes. While they aren’t allowed to dynamically switch out body elements during the event, they can change, for example, the angle of a wing flap. While these types of adjustments are usually performed without the benefit of last minute computer simulations, it’s certainly not a stretch to think that more complex modeling could be used for race day decisions in the not too distant future.

With all the merchandising money at stake (reportedly over a billion dollars per year) and with such a fierce level of competition, Formula One teams are likely to take advantage of any edge afforded by high end computing. Although most of the simulation work is currently focused on aerodynamic design, HPC software is also being used for FEA stress analysis and vehicle dynamics, as well as for validation of the final design.

And while computing hasn’t replaced wind tunnel testing, those days might not be too far off. Nevey recalls a number of instances last year when they were able to develop some aerodynamic components that went straight onto the car for initial testing, skipping the wind tunnel step entirely. As the engineers figure out how to do more sophisticated simulations, there should be even greater incentive to add more HPC resources into the mix. At Red Bull, there are already plans in the works to multiply their computational power by a factor of four.


As always, comments about HPCwire are welcomed and encouraged. Write to me, Michael Feldman, at

Share This