Can Innovation and Sustainability Coexist in Computer Aided Engineering?

How advances in semiconductors and simulation software help achieve both sustainability and innovation.

Altair has submitted this post.

Written by: Richard S. Yen, SVP, Global Industry Verticals, Altair

In modern manufacturing, CAE is essential. Engineers rely on these environmentally intensive simulations to boost product quality, improve time-to-market and reduce reliance on physical prototyping. However, manufacturers find themselves on the horns of a dilemma. On the one hand, they need to out-innovate competitors with better designs and shorter product cycles. On the other hand, they are challenged to build more sustainable, eco-friendly products while managing costs and minding their carbon footprints.

This article explores the tradeoff between innovation and sustainability. It discusses how advances in semiconductors and simulation software can help achieve both.

Sustainability Presents New Opportunities  

In business, sustainability has become more than just a buzzword; it is now an essential strategy for corporate success. A driving concern is the unsustainable levels of greenhouse gas (GHG) emissions causing climate change. Left unchecked, CO2 emissions are expected to lead to significant increases in average global temperatures. Many organizations have responded with public commitments around sustainability. These include pledges to accelerate a low-carbon energy transition and achieve net zero emissions by specific dates. According to an IDC FutureScape study, 60 percent  of organizations will have digital sustainability teams tasked with assessing, certifying and coordinating use of business and IT sustainability data and analytic platforms offered by ICT providers by 2025.[i]

Growing Simulation Requirements

Ironically, as manufacturers seek to reduce environmental impacts of their products, the need for simulation only increases. Lightweight products often mean honeycombed structures and more exotic materials, such as aluminum and carbon fiber composites, that are more difficult to model. Electric and hybrid vehicle manufacturers allocate significant resources to simulating battery technologies to ensure optimal performance and safety. They must also worry about strict emission and fuel efficiency regulations in many jurisdictions.

Simulation requirements are rapidly expanding beyond traditional structural analysis and computational fluid dynamics (CFD) workloads. For instance, products are increasingly embedded with smart connected devices, meaning engineers face new challenges related to spectrum management, electromagnetic compatibility (EMC) and electromagnetic interference (EMI). New AI techniques that enable autonomous and assistive driving features can be especially compute-intensive. Authors of deep learning models increasingly quote the environmental impact of model training. For example, the popular Stable Diffusion text-to-image model requires 150,000 compute hours to train and an associated carbon footprint of 11,250kg of CO2 equivalents.[ii] Not every model is this large, but manufacturers must deliver more computing capacity while managing costs.

Improving Data Center Efficiency

With state-of-the-art data center CPUs consuming up to 360 watts per socket or more, a single rack in a dense CAE facility can potentially consume ~40 kW. Powering such a rack for a year would consume ~350,000 kWh and can cost USD ~$56,000 — and this does not include cooling.[iii] As power becomes a higher input cost, energy efficiency is crucial for both sustainability and the bottom line.

Today, most CAE centers take an “all of the above” approach to improving efficiency. These include techniques such as intelligent simulation, schedulers to optimize resource use, improving power usage efficiency (PUE) and seeking renewable energy sources. Energy-efficient data centers start with energy-efficient processors, so selecting the right technology is critical.

Innovations in Silicon Help Point the Way

With energy efficiency being a top concern, processor manufacturers increasingly compete based on throughput per watt. Miniaturization helps improve both throughput and power efficiency. With smaller trace sizes, less resistance leads to lower power consumption. Shorter traces also have lower capacitance, meaning less power is required to charge/discharge lines. This leads to faster switching times, reduced propagation delays and higher clock rates.

For example, the latest AMD datacenter processors, 4th Gen AMD EPYC CPUs, are based on TSMC’s N5 (5nm) process. According to TSMC, this process delivers up to 20 percent better speed and up to a 40 percent power reduction compared to their N7 (7nm) process unveiled in 2018.[iv] While there are limits to how fast processors can be clocked, smaller, more power-efficient dies leave room for more cores, memory and I/O paths, and enable other chip-level innovations that help boost throughput. Today, multiphysics simulation software can scale to take full advantage of multicore processors, and 32-core processors are emerging as a “sweet spot” for many CAE tools.

CAE Energy Efficiency

AMD recently ran a series of CFD and finite element analysis (FEA) benchmarks to explore these tradeoffs between performance and energy efficiency. The tests were run using a variety of Altair solvers and standard models. As shown below, the latest processor technology delivered throughput improvements ranging between ~1.2x and ~2x compared to similar server configurations based on previous-generation processors.[v]

Relative throughput generational 32-core improvement using 4th Gen AMD EPYC processors. (Image courtesy of Altair.)

Relative throughput generational 32-core improvement using 4th Gen AMD EPYC processors. (Image courtesy of Altair.)

For CAE managers who labor to achieve single-digit gains in throughput and energy efficiency, improvements of up to 2x are dramatic. By provisioning servers with double the throughput, organizations can deliver the same simulation capacity with half the server footprint in some cases.[vi] This can translate into multiple savings opportunities, including reduced data center space, fewer racks and network drops, and lower power and cooling costs. 

Boost Energy Efficiency Without Compromise

Beyond the choice of processor, engineers can use advanced simulation techniques to help reduce data center costs and CO2 emissions further. Simulation-driven approaches can help engineers design more sustainable products that maximize performance while minimizing material usage. The annual Altair Enlighten Award honors advances in sustainability in the automotive industry and lightweight designs that boost mileage and result in more efficient products.

CAE users can also take advantage of other techniques to improve efficiency. These include leveraging workload managers with energy-aware scheduling features and employing software tools that facilitate “cloud bursting.”

For design center managers, the bottom line is that by selecting the right hardware and software, they can potentially have it both ways. They can expand their simulation capacity while simultaneously addressing sustainability commitments.

To learn more about how AMD and Altair are enabling sustainable computing, view the presentation by Kumaran Siva, Corporate Vice President, Software & Systems Business Development, AMD, at the recent Altair Future.Industry event.

[i] IDC FutureScape: Worldwide Sustainability 2022 Predictions,  Doc # US48312921, October 2021.

[ii] For details please see the “Environmental Impact” section runwayml/stable-diffusion-v1-5 · Hugging Face. Per the Hugging Face terms of service, this and other open repositories may be displayed, published, and reproduced by third parties.

[iii] 40 kW * 24 hours/day * 365 days per year = 350,400 kWh annually. Assuming an average rate of $.0.16 per kWh, 350,400 kWh * $0.16 per kWh = $56,064.

[iv] 5nm Technology – Taiwan Semiconductor Manufacturing Company Limited (

[v]  Altair Applications Technical Computing Summary Brief (

[vi] See the Altair Applications Technical Computing Summary Brief for details. In an Altair AcuSolve benchmark, a 2P server based on 4th Gen AMD EPYC 9374F 32-core processors delivered twice the throughput compared to a similar 2P server based on 3rd Gen AMD EPYC 75F3 32-core processors. This means that if achieving an aggregate throughput of “N” simulations per hour requires two racks of servers based on the 3rd Gen EPYC 75F3 processors, the same throughput can be achieved with a single rack and half the number of servers based on faster 4th Gen EPYC 9374F processors. Actual results will vary depending on the model simulated.