Perhaps a favorite aspect of my role on the AMD EPYC team is to meet brilliant people solving really complex challenges through continuous innovation. Some of these I get to capture and share via our AMD EPYC TechTalk podcast series. I recently had the pleasure to have one such discussion with Ms. Vicky Tsianika of Hexagon--a leader in digital simulation tools across many industries. Vicky is a leader in the team responsible for the MSC NASTRAN product line of sophisticated tools for Computer Aided Engineering (CAE) and simulating effects based on real world physics principles. You can listen to the full discussion here, but I thought I'd share some of the highlights of her insights here in a brief blog.
When you step into a vehicle -- whether it's a car, a plane or any other advanced piece of machinery -- there are countless conditions that could impact your ride. Is it too hot, outside or inside the vehicle? Is the wind slowing down your momentum, fuel efficiency or making excessive noise? Is the gravel on the road providing enough traction with your tires?
These may seem mundane, but these are examples of questions that designers and engineers have to think carefully about in the product development process. Experts in the automotive and aerospace industries are constantly testing and fine-tuning their designs, so they can withstand a wide variety of possible environmental changes. In the past, many of these scenarios were developed and tested with physical models–a time-consuming, inaccurate and costly process. Today, this process is aided by multiphysics simulation software, such as the tools provided by the digital reality company Hexagon. Hexagon develops a variety of design and simulation tools including the well-known MSC Nastran suite.
"I like to think of simulation as a virtual laboratory where engineers can actually implement their experiments and explore complex phenomena within a virtual environment," Hexagon's Vicky Tsianika explained.
Tsianika came to the US from Greece to pursue her Masters and PhD in mechanical engineering. Multiphysics, she explained, covers multiple physics, including fluid dynamics, acoustics, thermal and electrical phenomena–essentially critical ways to model complex real-world scenarios with higher fidelity.
"It can be really, really difficult to predict how each of these will impact the design, especially when they all affect each other," she said. "Even if you have a physical prototype, it's still very difficult to predict how all this --the natural forces the design is exposed to -- are going to affect its performance."
Of course, simulation and Finite Element Analysis (FEA) software is a game-changer in the automotive and aerospace industries. With predictive modeling that lets engineers assess how different designs and evolving conditions might impact their design, they can iterate much faster and earlier in the process. By simulating fluid dynamics and structural dynamics, organizations can uncover issues much more quickly and at lower cost.
Meanwhile, Tsianika said, AI is revolutionizing the capabilities of simulation software. Engineers can build digital twins that operate at lightning speed and deliver results in almost real time. They can explore data sets in a matter of seconds, potentially allowing them to discover the defective data sets or failure modes that even an experienced engineer may not have even thought about.
The compute demands in the simulation domain depend on the type of software and analysis being employed, Tsianika said. Traditionally, engineers have relied on powerful CPUs for their ability to handle complex algorithms and sequential tasks. When it comes to multiphysics simulations in general, there is a growing choice between using GPUs or CPUs depending on the specific requirements of the simulation.
Last December, Hexagon published a joint case study with AMD in order to deliver performance highlights of MSC Nastran, a structural analysis application, on a 4th Gen AMD EPYC processor-based system. Compared to the previous generation processor, performance improved 1.5x for a specific MSC Nastran workload.
While advances in AI and computational tools are advancing rapidly, engineers can't simply rely on software to test their designs, Tsianika stressed.
"No matter how sophisticated the multiphysics tools are, they still require human experience and the fundamental understanding of the underlying physics," she said. "There should be no expectation that there's something out there where you just push a button, and it gives you... a 100% accurate prediction. As engineers, we still have to be able to assess the right approach to the problem, create an accurate model and configure the simulation correctly."
At AMD we look forward to helping provide the hardware infrastructure to make the incredible tools from innovators such as Vicky Tsianika and Hexagon more efficient, more effective and more capable with each generation.
