Synopsys’ new simulation-driven framework promises to bring high‑fidelity computational fluid dynamics (CFD) and factory-scale digital twins into the operational tempo of the shop floor, but the bold performance claims that headline the announcement require disciplined verification before they can be used as procurement SLAs.
Background / Overview
Synopsys unveiled the framework at Microsoft Ignite as an open, GPU‑native architecture that stitches together accelerated physics, NVIDIA Omniverse/OpenUSD visualization libraries, Ansys Fluent GPU solvers, and Microsoft Azure cloud orchestration. The company and its partners say the solution can transform conventional CFD tasks — which commonly take hours — into near‑real‑time scenario comparisons that inform immediate production decisions. The initial, public deployment at packaging specialist Krones models entire bottle‑filling assembly lines as a physically accurate digital twin so operators and engineers can test variables such as bottle geometry, liquid viscosity and fill‑level without interrupting live operations. Partners named in the rollout include NVIDIA (Omniverse and CUDA‑X libraries), Microsoft Azure (cloud compute and Ansys Access on Azure), Ansys (Fluent), CADFEM Germany (solver tuning), and systems integrator SoftServe (platform integration). Independent materials and partner case studies corroborate the broad architecture — Ansys Fluent running on Azure with GPU acceleration and Omniverse/OpenUSD used as the interoperability and visualization layer — but they also document a notable discrepancy in the headline performance numbers that buyers must treat as material. The Synopsys announcement cites a reduction in CFD runtimes “from 3–4 hours to less than 5 minutes,” while an operational SoftServe case study tied to the Krones deployment reports fielded simulation cycles of roughly 30 minutes after tuning and integration. This gap matters for procurement, operations and engineering because it reflects differences in model fidelity, compute sizing and the use (or not) of reduced‑order/surrogate techniques.
What the framework actually is
Core components and the integration pattern
At a high level the framework unites four technical vectors:
- GPU‑native CFD: Ansys Fluent configured to use GPU acceleration and multi‑GPU scaling for fluid‑dynamic solves. This is the core physics engine for filling and flow problems.
- Accelerated physics / solver orchestration: Synopsys’ accelerated physics layer and cloud‑native orchestration that manage solver workflows, job distribution, and the integration of solver outputs into higher‑level pipelines.
- Visualization & interoperability: NVIDIA Omniverse libraries and OpenUSD (Universal Scene Description) as the scene/format and runtime for composing CAD/CAE geometry, telemetry and simulation outputs into a shared factory twin. This enables side‑by‑side scenario visualisation and collaborative review.
- Cloud HPC & deployment: Microsoft Azure as the elastic compute and orchestration layer — including Ansys Access on Microsoft Azure — enabling customers to run large GPU nodes, autoscale clusters, and host the digital twin dashboards.
Those components are combined into a scenario engine that sweeps process parameters, dispatches GPU solver jobs to Azure clusters, collects numerical results, and streams both analytics and 3D visualizations into an Omniverse digital twin for rapid comparison. SoftServe and CADFEM are credited with the domain tuning, solver configuration and delivery automation required to make that loop usable by production teams.
How OpenUSD / Omniverse is used
OpenUSD acts as the neutral scene and exchange format — a single container for CAD geometry, runtime metadata (conveyors, PLC signals), and CFD result overlays — so engineering, operations and analytics can remain synchronized. Omniverse supplies the real‑time rendering, scene composition and collaboration primitives for the twin. This choice lowers friction when integrating different CAE tools and makes it easier to surface simulation outputs to non‑CAE stakeholders.
The Krones deployment: what was delivered and why it matters
Krones, a global supplier of bottling and packaging systems, is the named early adopter whose filling lines were modeled and optimized using the framework. The deployment included:
- A factory‑scale digital twin of the filling line with parametric bottle and fluid models.
- CFD runs to test bottle shapes, fill levels and liquid properties, with scenario results presented via Omniverse visualizations and dashboards.
- Solver tuning by Ansys Apex Channel Partner CADFEM Germany and integration services by SoftServe to deliver a productionized pipeline.
SoftServe’s case study describes a two‑month delivery that produced
substantially faster simulation cycles compared with prior offline workflows, reporting simulation cycle times of roughly 30 minutes in their fielded Krones example. Those results are operationally meaningful: a 30‑minute turnaround brings simulation into the same shift cadence and enables
practical what‑if comparisons between engineering and operations without hours‑long waits. But the case study’s figure differs materially from the “under five minutes” claim in some press wiring, and that difference is central to the commercial evaluation.
Performance claims — verified facts and unresolved gaps
What the vendors say
Synopsys’ press materials and distributed press releases repeatedly present a headline reduction in CFD run times “from 3–4 hours to less than 5 minutes,” and quote senior partners — including NVIDIA and Microsoft executives — to frame the announcement as a milestone for real‑time, simulation‑driven operations. Those materials also emphasize OpenUSD interoperability and cloud scale as enabling factors.
What partner case studies show
SoftServe’s published Krones case study and integrator documentation present an actual, multi‑partner field delivery. That material details the stack (Ansys Fluent on Azure, Omniverse visualisation, Synopsys accelerated physics, CADFEM solver tuning) and reports simulation cycles around ~30 minutes in the fielded system — still a large improvement over hours‑long offline jobs, but not the sub‑5‑minute number. The SoftServe artifacts include time‑to‑deliver and practical engineering constraints, and are therefore an important, verifiable datapoint.
Why numbers can differ — a technical primer
- Mesh density and physics fidelity: CFD runtime grows (often nonlinearly) with mesh cell count and the complexity of physics models (multiphase flow, turbulence treatments, free‑surface capture). A full‑fidelity multiphase model that mirrors real fluids in a filling process can be dramatically slower than a reduced‑order or surrogate variant.
- Solver feature support on GPUs: Ansys Fluent has extended GPU support substantially (single/multi‑GPU, many solver options), but not every module or advanced physics is equally accelerated. The set of physics models used in a particular simulation will drive achievable speedup. Implementations may rely on hybrid CPU/GPU approaches or selective model simplification to reach aggressive runtimes.
- Compute footprint and cost: Achieving extreme wall‑clock speed often requires high‑end GPUs (A100/H100/MI300 series), fast storage and multi‑GPU interconnects — all of which increase per‑run cloud cost. Vendors sometimes report lab bench benchmarks that use optimized hardware profiles not representative of a typical production tenant.
- Use of surrogate models or physics‑informed ML: Near‑real‑time workflows sometimes use reduced‑order models, AI surrogates or hybrid physics/ML layers to produce very fast approximations. These deliver speed but require careful validation against full‑fidelity runs. Synopsys and partners reference AI‑assisted workflows as part of their narrative; if the five‑minute figure refers to a surrogate path rather than the full CFD solution, that distinction matters operationally.
Recommendation on claims
Until Synopsys or its partners publish reproducible benchmark runbooks — specifying mesh sizes, turbulence/multiphase models, exact solver flags, GPU types and counts, and the cost for the hardware profile used — the “under five minutes” claim should be treated as an aspirational or lab‑condition figure. The integrator case study (≈30 minutes) is a stronger operational datapoint and should be used as the conservative planning baseline for pilots and procurement.
Technical verification: what is supported today
Ansys Fluent’s GPU solver has matured into a production capability that supports single‑ and multi‑GPU workflows, a wide range of solver options (steady/transient, segregated/coupled solvers, species transport and several radiation models), and is certified on both NVIDIA and AMD server GPUs under specified driver and ROCm/CUDA versions. However, not all physics modules or workflows are equally accelerated; some advanced multiphase or particle methods can still be CPU‑bound or offer limited GPU acceleration. These are real engineering constraints that must be validated against your use case. Ansys Access on Microsoft Azure provides a production route to run Ansys products in a customer Azure tenancy with autoscaling, preconfigured images and job monitoring — a familiar cloud staging for enterprises that need to keep licensing and data governance under corporate control while leveraging Azure GPU SKUs. Omniverse and OpenUSD supply the visualization and interoperability layer, and NVIDIA has published industry blueprints and microservices that support industrial digital twins and physical AI workflows. Taken together the individual pieces are production‑grade; the engineering work required to make them operate as a repeatable pipeline is non‑trivial but achievable by experienced integrators.
Operational impact and business case
When the simulation‑to‑decision loop is tightened to minutes (even tens of minutes), the value potential is clear:
- Faster root cause and process tuning: Operators can test and compare scenarios during shift windows, reducing scrap and rework.
- Dynamic resource allocation: Rapid simulation allows better decisions about changeovers, recipe adjustments and preventive checks.
- Cross‑discipline feedback: Shared visual twins bridge engineering, operations and R&D with common artifacts and faster iteration cycles.
- Sustainability and waste reduction: Optimizing fill timing and valve settings can reduce product loss and energy usage — measurable KPIs for many CPG and food & beverage lines.
These are credible business impacts reported by vendors and early adopters; the economic model hinges on run‑rate simulation costs, integration overhead and the value of avoided waste or downtime. Realizing consistent ROI will require careful scoping and accurate cost modeling of GPU cloud hours.
Risks, caveats and governance
- Unverified headline metrics: As noted repeatedly, the gulf between “<5 minutes” and documented field times (~30 minutes) is material — require auditable benchmarks before committing to automated control decisions.
- Fidelity tradeoffs: Faster runs may rely on simplified physics or surrogates that introduce error bounds. Operational automation must include human‑in‑the‑loop thresholds and rollback conditions.
- Cloud costs and run‑rate economics: Frequent, minutes‑scale simulation at factory throughput requires sustained GPU hours; enterprises must model ongoing cloud spend, not only one‑time integration fees.
- Security and IP governance: Running detailed machine geometries and process models in the cloud raises IP risks. Enforce encryption, tenant isolation, identity controls, and secure CI/CD for models and datasets.
- Operational safety: Any automated control actions derived from simulation should pass safety interlocks, test harnesses and regulatory checks before being enacted on physical lines. Simulations can recommend changes; they must not be the single authority for safety‑critical actions.
How to evaluate the framework in a pilot — a practical checklist
- Define clear, measurable KPIs (e.g., percent reduction in waste, additional throughput, mean time to changeover) tied to the simulation outputs.
- Scope a bounded pilot: one filling line and a limited parameter sweep (e.g., 3 bottle geometries × 2 viscosity profiles).
- Request the integrator’s runbook: exact mesh size, solver flags, time steps, GPU SKU and cluster topology used for reported numbers. Demand the reproducible benchmark.
- Run side‑by‑side validation: full‑fidelity Fluent runs vs. accelerated pipeline vs. any surrogate outputs; quantify numerical differences and confidence intervals.
- Produce a cost model: compute cost per simulation at required cadence, storage and data egress, and platform engineering overhead.
- Define governance and safety: explicit human sign‑off limits, automatic rollback thresholds for any recommendations that would change actuator/PLC settings.
Strategic takeaways for IT and OT leaders
- The Synopsys framework represents a credible and pragmatic industry direction: bring CAE into the operational loop by combining GPU acceleration, cloud scale and standards‑based visualization (OpenUSD). The vendor ecosystem engaged (Ansys, NVIDIA, Microsoft, CADFEM, SoftServe) is well‑matched to the problem space.
- Buyers should insist on reproducible, auditable performance runbooks before treating sub‑5‑minute figures as contractual outcomes. The SoftServe/Krones field data (~30 minutes) is an important, verifiable baseline for realistic planning.
- Prepare for a non‑trivial integration lift: solver tuning, mesh and model management, cloud orchestration, and secure deployment pipelines are the operational work that turns vendor narratives into dependable capabilities. Systems integrators with proven Azure and Omniverse experience will be key to success.
Conclusion
Synopsys’ real‑time digital twin framework stitches mature, production‑grade pieces — Ansys Fluent GPU solvers, NVIDIA Omniverse/OpenUSD visualization, and Microsoft Azure cloud orchestration — into a compelling blueprint for bringing simulation into factory decision loops. The Krones deployment demonstrates practical gains and a credible path to tighter engineering‑operations feedback. At the same time, the variance between the headline “under five minutes” claim and documented, fielded runtimes (~30 minutes) is a critical purchasing and operational question that requires transparent, auditable benchmarking before organizations commit to scaled automation. With disciplined pilots, clear KPIs and governance that respects fidelity and safety tradeoffs, manufacturers can move from aspirational marketing to measurable operational value.
Source: IT Brief UK
Synopsys boosts manufacturing with real-time digital twin simulation