In 2022, Jon Peddie Research published a report on the significant role GPUs have played in the progress of computer-aided engineering (CAE). In the e-book “Accelerating and Advancing CAE Workflows,” we looked at GPU-accelerated applications from Altair, Ansys, Dassault Systèmes, and Hexagon. All of those companies were early to embrace the challenge of putting GPUs to work on legacy products built originally for CPUs.
In Part Two of our e-book series, “GPUs for CAE: New Platforms,“ we look at the potential benefits of GPU utilization in CAE simulations. GPUs can deliver over five times the throughput at a similar cost to CPUs, and they offer improved cost-efficiency and lower power consumption while maintaining the same throughput as CPUs.
Table of Contents
- Introduction: Where are we now?
- Evolved Computing
- A new era
- The best solution may be multiple solutions
- Artificial intelligence in CAE
- Summary
- Reference
Table of Figures
Figure 1. In 2020, Ansys and Peerless found that CAE professionals tended to use the resources closest at hand for CAE workflows. That picture is changing in 2023. When more powerful resources are available, users take advantage of them. (Source: Ansys’ Study on High-Performance Computing Usage for Engineering Simulation)
Figure 2. Automotive aerodynamics on CPU solver (left) and GPU solver (right). (Source: Ansys)
Figure 3. Speedup for different clusters and GPU server configurations. (Source: Ansys)
Figure 4. 0 GPUs indicates that the simulation was run with only CPUs. (Source: Microsoft)
Figure 5. Surface distribution of the pressure coefficient on the High Lift CRM at 14° angle of attack. (Source: Siemens)
Figure 6. Siemens says the turnaround time comparison between reference CPU (nine nodes, each with 64 cores, 576 total, clock speed of 2.4–3.3 GHz, 256 MB L3 cache) and GPU (6 GPU nodes, 24 NVIDIA A100s). Running on GPUs, an aerospace aerodynamic case (RANS, 110M cells) is completed in 40 minutes. (Source: Siemens)
Figure 7. High-fidelity CFD simulation with Fidelity CharLES of reacting compressible flow (left) and impact of GPU computing on high-fidelity CFD (right); Hardware: CPU c6a.32xlarge AMD Epyc Gen2, GPU p4d.24xlarge 8x NVIDIA A100. (Source: Cadence)
Figure 8. Altair’s aerodynamic and aero-acoustic workload results using NVIDIA H100 GPUs. (source Altair)
Figure 9. Digital Storm purpose-built CAE workstation employing NVIDIA AIBs. (Source: Digital Storm)
Figure 10. Boxx CAE mobile workstation. (Source: Boxx)
Figure 11. Comparison of CPU and GPU performance. (Source: Siemens)
Table of Tables
Table 1. Clock times for running the simulation, for both CPU and GPU configurations. (Source: Microsoft)
Table 2. CPU vs. GPU performance. (Source: Microsoft)
Introduction: Where are we now?
In 2022, Jon Peddie Research published a report on the significant role GPUs have played in the progress of computer-aided engineering (CAE). In the e-book “Accelerating and Advancing CAE Workflows,” we looked at GPU-accelerated applications from Altair, Ansys, Dassault Systèmes, and Hexagon. All of those companies were early to embrace the challenge of putting GPUs to work on legacy products built originally for CPUs.
Drawing from the conclusions of the previous e-book, GPUs provide some inherent advantages for running simulations:
- GPUs can achieve 5× or more throughput for the same cost of a CPU.
- They can achieve lower cost and power consumption for the same throughput as the CPU.
In JPR’s first e-book, software vendors told us that as GPU-accelerated CAE software applications have become available, their customers began using CAE simulation earlier in the design process. The users can run larger simulations faster on their desktop workstations, and as a result, they can optimize their designs with more iterations. That trajectory is picking up speed when more compute power from HPC and the cloud can be brought into play.
Customers who have been running their CAE simulations on legacy software and systems built for CPUs are recognizing the new opportunities opening through GPU usage, but this isn’t simply a transition. In the world of CAE and engineering, changes in methods and tools can entail expensive investments in new software and hardware. Also, those changes may bring about a transformation in where the work is done and who is doing it.
CAE apps can now run everywhere from workstations to HPCs on-site or off-premises, in the cloud, and in hybrid combinations forming a distributed and collaborative network.
In the past, simulation specialists were employed to create datasets and load them into the CAE programs, and then evaluate the results. With distributed computing, simplified CAE UIs, and some training, combined with hands-on experience, design engineers are taking on the work of more up-front simulations.
In some cases, this may be a lower-fidelity run, which can be done quickly on a desktop machine, allowing more iterations for better design efficiency, thanks to the speedup granted by GPUs. The more complex, multidisciplinary CAE studies may still be done by specialized and experienced analysts, but the scope of their capabilities is also expanding.
Download both e-books about GPUs for CAE:
Accelerating and Advancing Computer Aided Engineering Workflows