The automotive industry is now under pressure to develop and launch vehicles faster while also simultaneously managing greater product complexity. The transition to electric vehicles, increasing software content and the growing number of features and variants are stretching engineering resources and making development timelines increasingly difficult to compress. The traditional approach was to rely heavily on physical prototypes and testing, but it is no longer sufficient as manufacturers need to explore a much larger design space.
The development timeline illustrates the shift pretty clearly. Traditional vehicle development can take around 50 months from development through physical prototyping, testing and validation to start of production. Simulation-driven development has the potential to bring that timeline closer to 24 months, while the next generation of increasingly data-driven development approaches points toward further reductions.
The fundamental principle is pretty straightforward: move validation earlier in the development process, or shift left, so that more design decisions can be evaluated virtually before committing to physical hardware.
Moving more work into virtual development creates its own set of challenges for automotive engineering teams. High-fidelity computational fluid dynamics (CFD) models can involve hundreds of millions of computational cells and transient physics that require substantial computing resources.
Engineers need both speed and accuracy. And increasing one at the expense of the other will not solve the development problem. This is where advances in computing are becoming significant.
Speed that does not Compromise Engineering Confidence
The mobility industry has been exploring different ways to accelerate CFD, including improved numerical methods, parallel computing, faster pre-processing and GPU-based computing. Each has a key role to play. Faster meshing, for example, can directly increase the number of design variants that an engineer can evaluate in a day. Simulation has demonstrated Rapid Octree meshing by reducing the meshing time for a 300-million-cell model from approximately 10 hours to less than an hour, i.e., more than 10x the improvement. And such reductions have translated into the ability to assess two new or three modified designs in roughly the time previously required for one.
The next opportunity lies in the solver itself. Fluent's Native GPU Solver is designed specifically to take advantage of GPU computing for the CFD solution process. The distinction is important because not every GPU-enabled application uses the GPU in the same way.
In a conventional GPU-accelerated approach, portions of a computation may be offloaded from the CPU to a separate GPU backend. This can deliver meaningful acceleration, but the overall workflow can still be constrained by the parts of the application that remain CPU-dependent. Amdahl's law provides a useful way of understanding the limitation: if only part of a workload is accelerated, the unaccelerated portion eventually becomes the bottleneck.
A native GPU approach takes a path that is different. Rather than relying on CPU offloading, Fluent's GPU solver has been developed from the ground up to run natively on GPU hardware. The objective is to enable the complete solution process to take advantage of GPU parallelism, while also maintaining the robustness and accuracy expected from a production CFD solver. As automotive models become larger and more detailed, that distinction becomes increasingly pertinent.
From Faster Computation to More Engineering Iterations
The value of faster CFD is not merely that an individual simulation now finishes sooner. The greater benefit is actually what engineers can do with the time that they recover.
From external aerodynamics and under-hood thermal management to battery and electric-motor cooling, HVAC, aeroacoustics, liquid cooling and filling, and water management around vehicle sensors, automotive CFD covers a broad range of applications. And these applications frequently involve turbulence, heat transfer, radiation, multiphase flows, particles and rotating components.
For all these problems, speed can change the development question from ‘Can we run this simulation?’ to ‘How many design alternatives can we evaluate?’
The performance numbers show the true potential. Fluent benchmarks across automotive applications have shown speedups ranging from approximately 5–8X for an air-cooled electric motor and 6–9X for a large external-aerodynamics case, to 10–13X for an EGR valve and 10X for inverter filling. An auto water-wading case has also demonstrated a much larger 230X speedup under the tested configuration.
A large transient case also provided one more useful perspective**, as** a 250-million-cell, 10,000-timestep simulation that required approximately 52 hours on 512 CPU cores was completed in around 40 minutes using 64 GPUs.
Since vehicle development is inherently iterative, such improvements do matter. Aerodynamic surfaces, battery architectures, cooling systems, thermal management strategies and other components all necessitate engineers to explore alternatives and recognize trade-offs. A faster solver gives them more opportunities to do that before physical testing.
It also changes the economics of simulation. The computational workload is tied not just to hardware but also to how long that hardware has to operate. There are also substantial differences in electricity costs for the tested CPU and GPU configurations. This indicates that acceleration can have an impact beyond engineering turnaround time.
What Customers are Seeing
What is increasingly visible in automotive development is the practical value of GPU-accelerated CFD. Here’s how:
Volvo Cars, for example, has worked with Ansys and NVIDIA on aerodynamic simulations for the fully electric EX90. Using eight NVIDIA Blackwell GPUs for the solver, combined with CPU-based meshing, the overall simulation time was reduced from 24 hours to 6.5 hours. The solver itself achieved a 2.5X speed improvement compared with cost-equivalent hardware using 2,016 CPU cores. The significance goes beyond the headline speedup. For an electric vehicle, aerodynamic drag has a direct relationship with energy efficiency and driving range. The ability to perform more aerodynamic studies means engineers can investigate more design possibilities and make more informed decisions earlier in development.
Honda is another example at a substantially larger scale. At NVIDIA GTC 2026, Honda reported achieving 34X faster computation and a 38% reduction in cost using four NVIDIA GB200 GPUs compared with 1,920 cloud-based CPU cores while running high-fidelity, unsteady CFD through Ansys Fluent. Honda noted that GPU acceleration made large-scale CFD that had previously been impractical on CPUs feasible. For Honda, this is not simply a computing benchmark. Honda explained that the company was using the capability to accelerate its migration of CFD workloads from CPUs to GPUs while continuing to pursue safer, higher-quality products at an appropriate cost.
These examples reflect a broader movement in the automotive industry. At GTC 2026, NVIDIA highlighted Honda among manufacturers using GPU-accelerated industrial software and noted that such technologies are being used to replace lengthy CPU-based simulations with high-fidelity virtual testing and faster iteration.
The same direction is visible beyond conventional passenger vehicles. Major automotive manufacturers and motorsport organisations are looking at accelerated simulation to evaluate increasingly complex designs under tighter development constraints. For Formula 1 teams in particular, where aerodynamic performance is closely linked to competitive advantage and development cycles are continuous, the ability to obtain high-fidelity results faster can be particularly valuable.
The Next Opportunity: Not Just More Speed
The most interesting consequence of GPU acceleration may ultimately be what it makes possible. When simulation takes days, engineers naturally limit the number of cases they investigate. When the same analysis can be completed in hours, or in some cases minutes, the design process can become much more exploratory. Engineers can consider transient phenomena, larger models and more detailed physics that might previously have been difficult to accommodate within a development schedule.
This is already happening in automotive CFD. The Fluent GPU Solver supports applications spanning external aerodynamics, thermal management, liquid cooling and filling, turbulence, heat transfer, radiation, multiphase flows and rotating machinery.
The next step will be to combine this computational capability with other advances in engineering workflows. Faster pre-processing, GPU-native solvers, improved optimisation techniques and, increasingly, AI-based methods can complement one another. The purpose should not be to replace engineering judgment, but to give engineers the ability to explore more possibilities and extract greater value from every simulation.
Automotive development is ultimately a balance between engineering judgment, physics, cost and time. Faster computing does not change those fundamentals, but it does change the number of questions engineers can ask and answer within a development cycle.
And that is why GPU acceleration is becoming an enabler for a different way of developing vehicles: one in which high-fidelity virtual validation can move further left, design alternatives can be evaluated more extensively, and decisions can be made with greater confidence before physical prototypes are built. It is much more than a hardware upgrade for CFD.
The idea is not to simply make one simulation run faster, but to make the entire vehicle development process more capable of keeping pace with the demands that the next generation of mobility brings with it.
Pravin Nakod is the Director of Applications Engineering at Ansys. All views expressed are the author's own.