Fast Numerical Simulation (CAE)

Numerical simulations have been a cornerstone of industrial product and process development for decades. In particular, the Finite Element Method (FEM) has become the standard approach in structural mechanics. Due to the high computational demands involved, numerous efforts have been made to accelerate simulations—through High Performance Computing (HPC) or model order reduction, for example. However, these approaches often come at the expense of accuracy or require extensive precomputations.

At Fraunhofer IGD, we take a different approach: we develop and implement fast, accurate simulation methods based on FEM and fully optimized for Graphics Processing Units (GPUs). By leveraging the massive parallelism of modern GPUs, we achieve maximum computational performance. Our solution delivers results 10 to 100 times faster—without compromising accuracy. As a result, simulation times are dramatically reduced: processes that once took hours can now be completed in just a few minutes. Our applications have been developed specifically for structural mechanics and the simulation of distortion in 3D printing (e.g., laser-based powder bed fusion processes).

To unlock the full potential of these technologies in practice, the typical bottlenecks of conventional simulation approaches must be addressed. Our solutions offer tangible advantages in this regard.

These benefits can be demonstrated through our technologies and applications:

Despite significant advances in GPU performance and programmability in recent years, implementing simulation algorithms efficiently on GPUs still presents several practical challenges. With RISTRA, we address these challenges directly.

Limited GPU Utilization in Existing FEM Solutions

Numerical simulations—particularly those based on the Finite Element Method (FEM)—have long been a core tool in product and process development. However, conventional FEM solutions often utilize only a fraction of the computing power available in modern GPUs.

Because only selected calculation steps are offloaded to the GPU, intensive data transfers between the CPU and GPU are required. This creates a bottleneck that significantly limits potential performance gains.

Memory and Data Access Constraints on GPUs

To fully exploit GPU processing cores, the memory architecture presents a technical challenge. GPU memory is not as easily scalable as memory in CPU-based systems. In addition, conventional data structures often prevent efficient memory access and limit overall performance.

Our Solution: RISTRA

RISTRA, our GPU-accelerated structural mechanics solver, executes the entire simulation workflow directly on the GPU. Through specially developed algorithms and memory-efficient data structures, RISTRA fully exploits GPU parallelism and overcomes the memory and communication limitations of conventional approaches.

The result is exceptional acceleration factors—one to two orders of magnitude faster (10–100× speed-up) while maintaining high accuracy. This makes RISTRA ideally suited for applications in structural mechanics and 3D printing simulation.

The RISTRA product family can be flexibly integrated into existing development environments and offers significant benefits for both industrial companies and software providers: shorter development cycles, higher computational performance for complex simulations, reduced energy consumption, stronger competitive advantages through faster innovation

Get in touch with us. Our team at Fraunhofer IGD in Darmstadt will be happy to advise you on the application possibilities of the RISTRA product family and how accelerated simulation can benefit your development processes.