GPU‑accelerated BMW simulations
Siemens and Nvidia announced that GPU-based architectures can speed BMW aerodynamics simulations by up to 60%, enabling far larger compute throughput for design iterations. The collaboration highlights how GPUs are being used not just for ML but for massively parallel engineering simulations that shorten iteration loops. Faster simulation throughput can shift program timelines by allowing more design space to be explored before committing to tooling and fixtures. (x.com)
Before a car ever goes into a wind tunnel, engineers already know a lot about how air will wrap around its mirrors, wheels, and rear glass from software that solves millions of tiny airflow equations around a 3D model. Siemens’ Simcenter STAR-CCM+ is one of the main tools carmakers use for that kind of computational fluid dynamics work. (siemens.com) Those airflow calculations get chopped into a mesh, which is a digital net made of countless little cells covering the car and the air around it. NVIDIA says BMW runs many of these jobs every day, with single simulations reaching up to 100 million cells. (nvidia.com) For years, most of this work ran on central processing units, which are good at handling a wide mix of tasks but are not built to do the same math on huge numbers of cells at once. Graphics processing units were built to do many similar calculations in parallel, which makes them a better fit for this kind of physics problem. (nvidia.com) That is the shift Siemens and NVIDIA have been pushing together: move engineering simulation onto graphics processors, not just artificial intelligence training. Siemens said in June 2025 that BMW Group and Siemens had already achieved a 30 times speedup for transient aerodynamics runs on entire vehicle geometries using Simcenter STAR-CCM+ with NVIDIA Blackwell chips and NVIDIA CUDA-X libraries. (siemens.com) Transient aerodynamics means the software is not just checking a frozen snapshot of airflow, but following how the air changes over time as vortices form, move, and break apart around the car body. That matters because mirrors, wheel arches, and the back of the vehicle create messy wakes that can add drag and wind noise if engineers only look at a steady-state average. (nvidia.com) The new claim tied to BMW is not just raw speed on one run, but faster throughput across many runs. Siemens said in January 2026 that it plans to complete graphics-processor acceleration across its simulation portfolio so customers can run larger and more accurate simulations faster, which is the setup behind reports that BMW can cut aerodynamics turnaround times by as much as 60 percent. (siemens.com) That changes the design loop. If one airflow study finishes in hours instead of stretching deeper into the day, engineers can test more mirror shapes, wheel designs, grille openings, and underbody panels before a program locks down expensive tooling. (siemens.com) BMW is also using the same NVIDIA relationship far beyond airflow on the road car itself. In June 2025, BMW said its Virtual Factory program was scaling digital twins across all plants worldwide, with projected production-planning cost reductions of up to 30 percent. (press.bmwgroup.com) Put together, this is what the car industry has been trying to buy with bigger compute budgets: fewer guesses before metal gets cut. The interesting part of the Siemens-NVIDIA-BMW story is that graphics processors are becoming factory-floor and engineering tools at the same time, handling airflow physics for the car and digital-twin planning for the plant that will build it. (siemens.com)
