AUTOMOTIVE TESTING AND SIMULATION FOCUS: SUSTAINABILITY

Similarities between data and oil run deeper than their economic and societal value. The global appetite for information has never been stronger; a recent International Energy Agency study showed that internet traffic increased by a factor of 16 between 2010 and 2020, fueling a 9.4 times rise in data center workloads over the same period. With the same report claiming that these facilities consumed 200-250TWh of electricity, or 1% of the global total, in 2020, their environmental impact is coming under the spotlight too.

The need for data is even stronger in the automotive industry.

Jasmeet Singh, EVP and global head of manufacturing at Infosys, says high-performance computing (HPC) centers have become a vital tool as both product development and consumer-facing connected services become more complex. With the emergence of systems founded on artificial intelligence, he predicts even greater use of these facilities and a need for them to evolve in line with customer demands for efficiency.

"One of the primary bottlenecks in addressing energy efficiency is the limitation of traditional computing," the technologist comments. "The way to increase computational power is to increase the number of transistors, which in turn directly increases the power consumption. Another factor is poor design of high-performance computing facilities and their infrastructure."

Infosys is already working on solutions. Singh notes increased demand for HPC as a Service where users' needs are pooled on a shared cloud environment. Recent projects include transferring Daimler (now Mercedes-Benz Group AG and Daimler Truck) HPC workloads to the Lefdal Mine Datacentre in Norway as the car maker moves toward carbon neutrality by 2039. The facility is located in a mine and next to a fjord, offering a stable operating temperature without the need for cooling systems or water usage, which improves efficiency for the most energy-intensive processes.

Dr Richard Ahlfeld, CEO of Monolith AI, notes similar trends. He believes that sustainability has become a top three strategy for vehicle manufacturers, alongside autonomy and electrification. Development has become a data-intensive and often remote, cloud-based process, especially following Covid-19 lockdowns. The location of new data hubs and the energy supplied to them is a core consideration as the industry looks to drive down its environmental impact and operating costs.

"Putting those HPC centers somewhere where it's already cold, like in the north of Sweden, and having them powered with 100% renewable energy is a trend in automotive. That offers all the compute power that helps vehicles become lighter, faster, more efficient, and it can make that [process] climate neutral," Ahlfeld says.

Ahlfeld cites BMW as an example. In 2020, it signed a six-year contract with EcoDataCenter, using 4MW of HPC capacity at the company's facility in Falun, Sweden with sustainability criteria as a priority. The data center uses 100% hydropower and excess heat from the server halls is fed to a nearby power plant, which produces renewable biofuel pellets. It offers high power usage effectiveness [PUE] of every 1.15kWh supplied to the center, lkWh is directly used for?data transfer and processing – and the heat generated cuts 700-800 hours of fossil fuel operations at the power plant, resulting in what's claimed to be the first climate-positive facility of its kind.

MONEY MAKER

Suppliers are recognizing the business opportunity in going green. "If you look at the three big data center providers Microsoft Azure, Amazon AWS and Google Cloud - Google Cloud is the worst performing," comments Ahlfeld. "It has recently announced that it's going to go 100% renewable energy, and I think that's a very competitive value proposition. If your cloud platform provider is moving to that, its another interesting sign that there is a demand from the market."

Walt Hearn, VP of customer excellence at Ansys, sees HPC as a natural focus point as manufacturers push for more sustainable processes. He points out that energy and data- intensive simulations for Volkswagen and Audi cars, such as crash and wind tunnel testing, are now being handled by Green Mountain's hydropower data center in Rjukan, Norway, and autonomous driving is expanding the need for such facilities. A deep learning model could generate 5113/hr for eight hours a day, he says, equating to 20PB/wk for a fleet of 100 vehicles (equivalent to uploading 100 billion Facebook photos).

However, improvements at a chip level are just as important. "The enhanced CPU architectures offered by AMD in its latest third-generation EPYC 7003 family of processors with AMD 3D V-Cache technology can improve simulation throughput performance by as much as 80% for several key CAE workloads," Hearn explains.

"This increased performance can also enhance sustainability initiatives by reducing the number of required HPC servers and the use of energy to perform larger numbers of simulations. An end user running CFD simulations could cut the number of servers required in half by migrating from a competitive emiironment to AMD EPYC 7003, while cutting power consumption in half."

Singh also sees potential in new computer architectures, highlighting the growing interest in neuromorphic computing, which mimics the way neurons in the human brain learn.

He says that as well as expanding the capabilities of AI, it offers significantly improved energy efficiency compared with previous architectures. Intel's global research community includes more than 100 members working on applications for this technology.

MORE FOR LESS

Monolith AI believes some of the incoming challenges can be addressed by smarter use of data. Ahlfeld says customers realize that there's a shortage of data scientists who can carry out the most detailed simulations, and an associated need for more efficient algorithms that use less data but offer a deeper insight.

The company is using AI to predict a wide variety of outcomes based on engineers' previous test data, which it says can reduce the number of simulations required for hard-to-improve features such as aerodynamics. It's claimed that this has offered 70% reductions in track time and 45% cost savings, while also simplifying processes and making them accessible to a bigger pool of engineers.

“ By using hardware more effectively, customers can simulate larger, more complex models” Walt Hearn, VP of customer exellence, Ansys.

DEEP SEA DATA

Microsoft began exploring alternatives to land-based data centers in 2014. Project Natick deployed its first prototype under-sea system in 2015, and followed it up with a second in 2018, assessing an environment the company believes could offer some inherent operational benefits.

Phase Two finished in 2020 with the retrieval of a data center the size of a shipping container from the North Sea near Scotland's Orkney Islands after two years underwater. The unit is dry nitrogen-filled and unmanned, and features 12 racks and 864 servers powered by 100% renewable electricity from wind, solar, tidal and wave.

The company is optimistic about future deployments. With no need for cooling, it achieved a PUE of 1.07 and without the water required to cool land-based systems. Failure rates were eight times lower than on land and, with half of the world population living close to the ocean, similar centers could be more local, which would reduce latency too.

"One of the tricks available with machine learning and AI is learning from everything that you've done in order to make predictions. The industry wasn't built like that — people never thought, 'If they just save my data today, that might become useful in the future', but that is definitely a mindset," he says.

"There is a big opportunity to reduce physical testing. If I take 15,000 examples from the past, maybe they can tell me what to do, because reality doesn't change that much."

Hearn has a similar view, adding that other sectors can learn from automotive's use of HPC during development, especially for complex and energy-intensive tasks such as crash testing and aerodynamics. And a more efficient use of the computing hardware benefits engineering productivity, and in turn catalyzes further sustainability improvements.

"It means they can basically consider more design ideas and make efficient prochict development decisions based on an enhanced understanding of performance trade-offs," he comments. "By using hardware more effectively, customers can simulate larger, more complex models so that more accurate design decisions can be made throughout the development process."