Lenovo says it has expanded the high-performance computing (HPC) infrastructure used by Ducati Corse to develop its 2027 Desmosedici GP MotoGP motorcycle. The company reports that the system cuts 3D engine computational fluid dynamics (CFD) simulation time by 50% and structural-analysis time by 30%.

The infrastructure is being used alongside continued development of Ducati’s 2026 motorcycle. Lenovo says the 2027 MotoGP technical-regulation changes require major redesign work, including on the engine and aerodynamics, while the team continues its current-season programme.

The figures come from a Lenovo manufacturer announcement published on 9 September 2026, rather than an independent benchmark report.

Contents

What changed

The expanded infrastructure provides more than 108 TFLOPS of combined theoretical double-precision computing performance, according to Lenovo. TFLOPS measures the theoretical number of floating-point operations a system can perform per second; double precision uses a higher numerical precision commonly relevant to scientific and engineering workloads.

Compared with Ducati Corse’s previous generation of compute nodes, Lenovo says the expansion roughly doubled available memory and added approximately 48% more processing power. The company says this allows engineers to run several engine and aerodynamic simulations at the same time and examine a wider range of virtual design configurations.

One important application is development of an all-new 850cc engine. The release does not provide the specific processors, accelerators, servers, storage or networking hardware used in the system.

The practical change is therefore not simply a larger headline performance figure. More memory and additional compute capacity can allow engineering teams to run more workloads concurrently, reducing the time spent waiting for one simulation to finish before starting another.

How the computing is used

CFD uses numerical computation to estimate how fluids move around or through a design. In motorcycle development, such models can estimate quantities including pressure, aerodynamic forces and heat transfer around the bike and its components.

Structural analysis addresses a different part of the engineering problem. It uses computational models to estimate how parts respond to loads, including stress, deformation and possible failure conditions. Engine, chassis and aerodynamic components must be evaluated not only for their performance but also for whether they can withstand expected forces.

Lenovo says Ducati is using the expanded capacity to run engine and aerodynamic simulations simultaneously. This can support a broader virtual design search: engineers can compare more shapes, configurations or component choices within a given period instead of evaluating them sequentially.

The release also frames simulation and predictive engineering as particularly important because opportunities for on-track testing are limited by MotoGP’s concessions system. It says the more capable infrastructure supports increasingly sophisticated models intended to represent real-world aerodynamic conditions more accurately. The source does not provide the detailed rules of that system or describe how the models are validated against physical measurements.

The team is working on two development demands at once: finalising the next-generation motorcycle for 2027 while continuing to optimise the 2026 bike. Faster simulation can help manage that workload, but the benefit depends on more than raw computing capacity. Model quality, software efficiency, input data and validation all affect whether additional simulations produce useful engineering information.

What the performance figures show—and do not show

Lenovo reports a 50% reduction in 3D engine CFD simulation time and a 30% reduction in structural-analysis time. These are useful indicators of the claimed change in workflow speed, but the release does not identify the workloads, problem sizes, software versions, baseline hardware or measurement method behind the comparisons.

It also does not provide repeated measurements, experimental controls or independent verification. The 108 TFLOPS figure is theoretical double-precision throughput and should not be treated as equivalent to end-to-end CFD performance, simulation accuracy or motorcycle performance.

Nor does the announcement establish that the computing expansion has directly produced lower costs, a shorter overall development schedule, improved model accuracy or better race results. Faster virtual iteration could allow Ducati engineers to assess more alternatives before building physical components, but that outcome is not quantified in the source.

Simulation remains an approximation of physical behaviour. Its value depends on the assumptions built into the model and on comparison with physical testing or other measurements. No evidence in the announcement shows that the 2027 Desmosedici GP has been completed, raced or demonstrated superior on-track performance.

What to watch

Further technical details would make the performance claims easier to assess. These include the compute-node hardware, accelerator configuration, software stack, simulation sizes and the precise baselines used for the reported time reductions.

The more meaningful engineering outcomes will also emerge later: whether the system changes the number of virtual configurations Ducati can assess, affects prototype requirements or shortens development milestones. Ultimately, the final 2027 motorcycle’s physical testing and race performance will show how effectively the additional simulation capacity translated into a working design.

Sources