How HPC supports 'continuous integration of new ideas' for optimizing Formula 1 car design

Alfa_Romeo_C38_Antonio_Giovinazzi_ITA_ALFA_ROMEO_RACING_(46388731925).jpg

The next BriefingsDirect extreme use-case for high-performance computing (HPC) examines how the strictly governed redesign of Formula 1 race cars relies on data center innovation to coax out the best in fluid dynamics analysis and refinement.

We’ll now learn how Alfa Romeo Racing (formerly Alfa Romeo Sauber F1 Team) in Hinwil, Switzerland leverages the latest in IT to bring hard-to-find but momentous design improvements — from simulation, to wind tunnel, to test track, and ultimately, to victory. The goal: To produce cars that are glued to the asphalt and best slice through the air.

<br />

Listen to the podcast. Find it on iTunes. Read a full transcript or download a copy.

Here to describe the challenges and solutions from the compute-intensive design of Formula 1 cars is Francesco Del Citto, Head of Computational Fluid Dynamics Methodology for Alfa Romeo Racing, and Peter Widmer, Worldwide Category Manager for Moonshot/Edgeline and Internet of Things (IoT) at Hewlett Packard Enterprise (HPE). The discussion is moderated by Dana Gardner, Principal Analyst at Interarbor Solutions.

Here are some excerpts:

Gardner: Why does Alfa Romeo Racing need to prepare for another car design?

Del Citto: Effectively, it’s a continuous design process. We never stop, especially on the aerodynamic side. And what every Formula 1 team does is dictated by each race season and by the specific planning and concept of your car in terms of performance.

For Formula 1 racing, the most important and discriminating factor in terms of performance is aerodynamics. Every Formula 1 team puts a lot of effort in designing the aerodynamic shape of their cars. That includes for brake cooling, engine cooling, and everything else. So all the airflow around and inside of the car is meticulously simulated to extract the maximum performance.

Gardner: This therefore becomes as much an engineering competition as it is a racing competition.

Engineered to race

Del Citto: Actually, it’s both. On the track, it’s clearly a racing competition between drivers and teams. But before you ever get to the track, it is an engineering competition in which the engineers both design the cars as well as the methods used to design the cars. Each Formula 1 team has its own closely guarded methodologies and processes – and they are each unique.

Gardner: When I first heard about fluid dynamics and aerodynamic optimization for cars, I was thinking primarily about reduction of friction. But this is about a lot more, such as the cooling but also making the car behave like a reverse airplane wing.

Tell us why the aerodynamic impacts are much more complicated than people might have appreciated.

Del Citto: It is very complicated. Most of the speed and lap-time reductions you gain are not on the straightaways. You gain over your competitors in how the car behaves in the corners. If you can increase the force of the air acting on the car — to push the car down onto the ground — then you have more force preventing the car from moving out of line in the corners.

Why use the force of the air? Because it is free. It doesn’t come with any extra weight. But it is difficult to gain such extra inertial control forces. You must generate them in an efficient way, without being penalized too much from friction.

Learn How High-Density HPC

Doubles Throughput

While Slashing Energy Use

It’s also difficult to generate such forces without breaking the rules, because there are rules. There are limits for designing the shapes of the car. You cannot do whatever you want. Still, within these rules, you have to try to extract the maximum benefits.

The force the car generates is called downforce, which is the opposite of lift force from the design of an airplane. The airplane has wings designed to lift. The racing car is designed to be pushed down to the ground. The more you can push to the ground, the more grip you have between the tires and the asphalt and the faster you can go in the corners before the friction gives up and you just slide.

Gardner: And how fast do these cars go nowadays?

Del Citto: They are very fast on the straight, around 360-370 km/hour (224-230 mph), especially in Mexico City, where the air is thin due to the altitude. You have less resistance and they have a very long straight there, so this is where you get the maximum speeds.

But what is really impressive is the corner speed. In the corners you can now have a side acceleration force that is four to five times the force of gravity. It’s like being in a jet fighter plane. It’s really, really high.

Widmer: They wear their security belts not only to hold them in in case of an accident, but also for when they brake and steer. Otherwise, they could be catapulted out of the car because the forces are close to 5G. The efficiency of the car is really impressive, not only from the acceleration or high speeds. The other invisible forces also differentiate a Formula 1 car from a street car.

Gardner: Peter, because this is an engineering competition, we know the simulations result in impactful improvements. And that then falls back on the performance of the data center and its level of innovation. Why is the high-performance computing environment such an essential part of the Formula 1 team?

Widmer: Finding tens of thousands of a second on the racetrack, where a lap time can be one minute or less, pushes the design of the cars to the extreme edge. To find that best design solution requires computer-aided design (CAD) guidance — and that’s where the data center plays an important part.

Those computational fluid dynamics (CFD) simulations take place in the data center. That’s why we are so happy to work together with Alfa Romeo Racing as a technology partner.

Gardner: Francesco, do you have constraints on what you can do with the computers as well as what you can do with the cars?

Limits to compute for cars

Del Citto: Yes, there are limits in all aspect of the car, design, and especially in the aerodynamic research. That’s because aerodynamics is where you can extract more performance — but it’s where you can spend more money as well.

The Formula 1 governing body, the FIA, a few years ago put in place ways of controlling the money spent for aerodynamic research. So instead of putting on a budget cap, they decided to put a limit on the resources you can use. The resources are both the wind tunnel and the computational fluid dynamics. It’s a tradeoff between the two. The more wind tunnel you use, the less computational power you can use, and vice versa. So each team has its sweet spot, depending on their strategy.

You have restrictions in how much computational capacity you can use to solve your simulations. You can do a lot of post-processing and pre-processing, but you cannot extract too much from that. The solving part, in which it tells you the performance results of the new car design, is what is limited.

Gardner: Peter, how does that translate into an HPE HPC equation? How do you continuously innovate to get the most from the data center, but without breaking the rules?

Widmer: We work with a competency center on the HPC to determine the right combination of CPU, throughput, and whatever it takes to get the end results, which are limited by the regulations.

We are very open on the platform requirements for not only Alfa Romeo Racing, but for all of the teams, and that’s based on the most efficient combination of CPU, memory, networking, and other infrastructure so that we can offer the CFD use-case.

Learn How High-Density HPC

Doubles Throughput

While Slashing Energy Use

It takes know-how about how to tune the CPUs, about the specifics of the CFD applications, and knowledge of the regulations formula which then leads us to get that success in CFD for Formula 1.

Gardner: Let’s hear more about that recipe for success.

Memory makes the difference

Widmer: It’s an Intel Skylake CPU, which includes graphic cards onboard. That obviously is not used for the CFD use-case, but the memory we do use as a level-four memory cache. That then provides us extra performance, which is not coming from the CPU, which is regulated. Due to the high-density packaging of the HPE Moonshot solution — where we can put 45 compute notes in a 4.30 rack chassis — this is quite compact. And it’s just topped out at about 5,000-plus cores.

Del Citto: Yes, 5,760 cores. As Peter was saying before, the key factor here is the software. There are three main CFD software applications used by all the Formula 1 teams.

The main limitation for this kind of software is always the memory bandwidth, not the computational power. It’s not about the clock speed frequency. The main limitation is the memory bandwidth. This is why the four-level cache gives the extra performance, even compared to a higher spec Intel server CPU. The lower spec with low energy use CPU version gives us the extra performance we need because of the extra memory cache.

Gardner: And this isn’t some workload you can get off of a public cloud. You need to have this on-premises?

Del Citto: That’s right. The HPC facility is completely owned and run by us for the Formula 1 team. It’s used for research and even for track analysis data. We use it for multiple purposes, but it’s fully dedicated to the team.

It is not in the cloud. We have designed a building where we have a lot of electricity and cooling capacity requirements. Consider that the wind tunnel fan — only the fan – uses 3 megawatts. We need to have a lot of electricity there.

Gardner: Do you use the wind tunnel to cool the data center?

Del Citto: Sort of. We use the same water to cool the wind tunnel and the data center. But the wind tunnel has to be cooled because you need the air at a constant temperature to have consistent tests.

Gardner: And Peter, this configuration that HPE has put together isn’t just a one-off. You’re providing the basic Moonshot design for other Formula 1 teams as well?

A winning platform

Widmer: Yes, the solution and fit-for-regulations design was so compelling that we managed to get 6 out of 10 teams to use the platform. We can say that at least the first three teams are on our customer list. Maybe the other ones will come to us as well, but who knows?

We are proud that we can deliver a platform to a sport known for such heavy competition and that is very technology-oriented. It’s not comparable to any other sport because you must consistently evolve, develop, and build new stuff. The evolution never stops in Formula 1 racing.

For a vendor like HPE, it’s really a very nice environment. If they have a new idea that can give a team a small competitive advantage, we can help them do it. And that’s been the case for 10 years now.

Let’s figure out how much faster we can go, and then let’s go for it. These teams are literally open-minded to new solutions, and they are eager to learn about what’s coming down the street in technology and how could we get some benefits out of it. So that’s really the nice story around it.

These teams are literally open-minded to new solutions, and they are eager to learn about what’s coming down the street in technology and how they could get benefits out of it. That’s the nice story around it.

Gardner: Francesco, you mentioned this is a continuous journey. You are always looking for new improvements, and always redesigning.

Now that you have a sophisticated HPC environment for CFD and simulations, what about taking advantage of HPC data center for data analysis? For using artificial intelligence (AI) and machine learning (ML)?

Is that the next stage you can go to with these powerful applications? Do you further combine the data analysis and CFD to push the performance needle even further?

<br />

Del Citto: We generate tons of data — from experiments, the wind tunnel, the CFD side, and from the track. The cars are full of sensors. During a practice run, there are hundreds of pressure sensors around the car. In the wind tunnel, there are 700 sensors constantly running. So, as you can imagine, we have accumulated a lot of data.

Now, the natural step will be how we can use it. Yes, this is something everyone is considering. I don’t know where this will bring us. There is nothing else I can comment on at the moment.

Gardner: If they can put rules around the extent to which you can use a data center for AI, for example, it could be very powerful.

Del Citto: It could be very powerful, yes. You are suggesting something to the rule-makers now. Obviously, we have to work with what we have now and see what will come next. We don’t know yet, but this is something we are keeping our eyes on, yes.