Formula 1, Rob Smedley explained, is maybe the biggest racing spectacle in the world, with five hundred million fans tuning in for every race. Smedley, a chief engineer with Formula 1’s performance engineering and analysis team, took the stage at Amazon Web Services (AWS) re:Invent 2019 to describe how HPC is making the next generation of Formula 1 possible.
“There’s a key strategic moment coming up for us – an opportunity in 2021, where we’re going to make seismic shifts and changes in the technical regulations in the sport overall,” Smedley said. A while back, they approached fans asking what they wanted out of the next generation of the sport. Across the board, they wanted more wheel-to-wheel racing, with the cars mere inches away from each other.
Aerodynamics, of course, make that difficult. As a Formula 1 car’s speed increases, the downforce goes up, pushing the car into the ground – “exactly the opposite” of an airplane taking off. But those forces also create a massive wake, drastically slowing any cars following close behind (“Wake bad,” Smedley said). Following one second behind a car can decrease downforce by 30 percent; at half a second behind (roughly the distance required for wheel-to-wheel racing) that downforce decrease becomes 40 percent.
So to minimize wake and maximize downforce for their next generation of cars, Formula 1 turned to computational fluid dynamics (CFD). Formula 1, Smedley explained, has been using CFD to create “virtual cars in virtual airspaces” for about 20 years with increasing sophistication and insight into the physics of airflow. And CFD, of course, required high-performance computing.
“CFD has always been a supercomputer problem,” Smedley said. “It’s always been an HPC problem.” Formula 1’s beginnings in HPC were humble; early efforts were just twenty-odd CAD terminals daisy-chained together overnight and disconnected when the designers returning in the mornings. “Today,” Smedley said, “we use about 200 cores … and it’s about two and a half thousand times the computational power.”
Formula 1 built a two-car simulation to visualize the wake effect, hoping to iterate on their next-generation car designs in ways that reduced the wake as much as possible. Still, the scope of the computing was challenging. “If you take a single-car CFD run and you take the most powerful home computer on the market today,” Smedley said, “it would take around 14 days.” Even with Formula 1’s 200-core system, individual runs took four full days – nowhere near the capability that their ambitious goal required.
So Formula 1 partnered with AWS to run their simulations on the cloud, instead. Smedley ran briefly through the numbers: 7,300 cores solving 14 cases simultaneously, with a total of 2.7 billion cells to be solved in the CFD models. “Partnering with AWS and their cloud HPC services,” Smedley said, “the first iteration for a two-car simulation, we got down to eleven and a half hours. We’re now at less than eight hours to do a two-car simulation.”
“And what that’s done is given you this behind me: the 2021 Challenger,” Smedley said. “This is what we hope is going to change Formula 1, and we are hopeful that it’s exactly what the fans are asking for.” He explained that the simulations helped them develop a new design with “up-washing” wake that would travel over (rather than onto) cars in hot pursuit.
Indeed, the simulation results are striking: at one second behind another car, the downforce reduction changed from 30 percent to 5 percent; at half a second behind, from 40 percent to just 7 percent.
“It’s exactly what we’ve been searching for,” Smedley said.