Cross Industries
Schneider AVEVA Immersive Twin Helps AI Factories Get More Compute per Watt
AVEVA
Objective
Designing AI factories is complex, and optimizing their power use and efficiency is crucial. AVEVA, part of Schneider Electric, is bringing its industrial engineering software to AI factories with NVIDIA Omniverse™ and NVIDIA DSX Sim™, part of the NVIDIA DSX™ AI Factory platform. Building on the NVIDIA Omniverse DSX Blueprint and SimReady equipment assets, AVEVA connects equipment data with cooling, power, and process solvers. Its application supports DSX Sim workflows by helping engineering teams evaluate IT and facility configurations together before making physical changes and find opportunities to free power for AI.
Customer
Nscale, Start Campus
Partner
AVEVA, part of Schneider Electric
Topic
Simulation / Modeling / Design
Products
Key Takeaways
- More power for AI workloads: Using AVEVA’s digital twin application, engineers can evaluate cooling, power, and process scenarios to identify opportunities to make more of the facility’s power budget available for AI compute.
- Lower-risk equipment changes: Teams can evaluate IT and facility equipment changes in a digital twin developed on OpenUSD and NVIDIA Omniverse, using SimReady assets, before making physical changes.
- Engineering analysis in one place: The twin shares equipment data with cooling, power, and process solvers and is designed to accommodate third-party solvers of the customer’s choice.
Bringing Cooling, Power, and Process Simulation Into One Digital Twin
Cooling, power, and process analysis often involves specialized solvers, and moving data between separate tools makes handoffs harder. Engineering teams may also have solvers they already rely on. Schneider Electric EcoStruxure IT Design CFD, for example, models airflow inside the data hall; extending the analysis outside the building requires additional capabilities.
AVEVA's Immersive Twin ties these tools together. It maintains an ontology—a map linking equipment in the OpenUSD scene with engineering solvers and metering data—and combines that information in the twin. On the design side, the solution builds on existing engineering workflows and tools, integrating with authoring solutions such as AVEVA Unified Engineering to generate the OpenUSD scene. The ontology and data model can be defined and traced as a single source of truth within the solution or managed via connections to external systems such as AVEVA Asset Information Management.
On the simulation side, the solution connects to Schneider Electric EcoStruxure IT Design CFD, ETAP for power simulation, and AVEVA Process Simulation. The architecture is also designed to support third-party solvers selected by the customer. Data connectivity is similarly extensible: through NVIDIA DSX Exchange™, teams can integrate a wide range of data sources and then store, curate, and align them in systems such as AVEVA PI.
The application is rendered on NVIDIA RTX GPUs and streamed to a web browser using Omniverse's WebRTC streaming capability. It can be deployed and scaled on premises or on cloud infrastructure with NVIDIA RTX™ GPUs.
To speed development while maintaining quality, AVEVA paired AI-assisted coding with a modular OpenUSD architecture, CI/CD, code reviews, and automated tests. The company credits the NVIDIA Omniverse DSX Blueprint with significantly accelerating its effort to address a new market. The team used AI tools to scaffold Omniverse integration code and generate initial test cases. The same approach helped the team identify some of the viewport's heavy computations and offload them to the GPU. AVEVA reports that these changes significantly improved viewport responsiveness and overall performance. AVEVA says AI assistance also helps the team turn customer feedback into rapid prototypes and quick validation cycles. AVEVA progressed from its initial workshop to a working prototype in four weeks.
In just six months, an AVEVA Tiger Team delivered the first end-to-end Immersive Twin MVP. It combines OpenUSD scene composition, an extensible integration and orchestration hub, and six representative solver and data-source integrations. Using High Velocity Engineering practices, including AI-assisted development, the team rapidly turned an ambitious vision into a working, integrated solution and established the architectural foundation for enterprise adoption, with traceability, extensibility, and a clear path to production.
“With NVIDIA Omniverse and OpenUSD, we help customers manage the full lifecycle of AI factory projects, from design to optimization, by unifying equipment data and engineering simulation in a single, immersive digital twin. Teams can evaluate IT, power, and cooling together with their preferred solvers, validate changes before deployment, and drastically reduce project timelines.”
Arti Garg
Chief Technologist, AVEVA
AI factory digital twin showing computational fluid dynamics (CFD) simulation, with color-coded airflow streamlines around server racks to evaluate cooling performance.
Planning IT and Facility Upgrades With SimReady Assets
An IT upgrade can change a facility’s power and cooling requirements, so teams need to evaluate IT and facility equipment together. Information about individual servers sits in separate systems, including data center infrastructure management (DCIM) tools and cluster managers, making it harder to connect IT activity with the physical equipment that supports it.
AVEVA’s digital twin brings IT and facility equipment into one model. For a pilot with Nscale and Start Campus at a data center in Portugal, the team imported the site’s building design and populated it with SimReady assets: 3D equipment models that carry engineering properties. The application uses Omniverse Fabric to populate assets in the scene. AVEVA’s software automatically reads their available properties from cloud-stored validated asset files and fills in rack and component lists. The application is designed to accommodate SimReady assets from multiple equipment vendors.
Engineers can change rack configurations, edit equipment parameters, and save and track scenarios. AVEVA is building scenarios where engineers model an IT upgrade, evaluate its effect on cooling requirements, and test whether a facility equipment change, such as a new coolant distribution unit (CDU), can meet those requirements.
Each server in the model is also structured to link to metadata and real-time data from IT systems. This mapping is designed to help IT users locate running jobs in the same model facility engineers use to plan upgrades.
For an IT refresh that increases rack power density, AVEVA’s digital twin lets teams test rack and facility changes in one traceable model and run power and cooling simulations that leverage SimReady engineering attributes for rapid feedback. By auto-populating the scene from validated SimReady assets, teams can cut setup from days to hours and iterate each new configuration in minutes to a few hours, versus multi-day cycles in traditional, disconnected workflows.
Those equipment changes also affect the facility’s power and cooling requirements.
AVEVA
Optimizing AI Factory Power Use With Cooling Simulation
Every AI factory operates within a set power envelope. Inefficient cooling design can divert power from AI compute, limiting GPU utilization and overall efficiency. The challenge is reducing cooling power without pushing equipment past its temperature targets.
AVEVA addresses this with a cooling optimizer built into its digital twin application, developed with NVIDIA Omniverse libraries and OpenUSD. The application uses Omniverse Kit SDK for its 3D interface. Working from the data hall’s current layout, the optimizer holds IT load constant and adjusts coolant temperature at the CDUs in simulation. Engineers can see how temperatures respond against their targets in an animated view, review detailed dashboards, and apply the same analysis to training, inference, and burst workload scenarios.
The optimizer is designed to help engineers identify cooling settings that reduce required fan wall and CDU power while meeting temperature targets, potentially making more power available for AI workloads within the same facility power envelope.
When IT equipment changes, the demands on the cooling infrastructure change too, making equipment configuration another important part of the engineering workflow.
AVEVA
Extending AI Factory Digital Twins With More Solvers and AI Agents
As the solution moves beyond its initial development team, AVEVA plans to add more third-party solvers and analytic tools. It also anticipates using AI agents to support user interaction and analytics across connected solvers. These extensions would build on its OpenUSD-based digital twin, which brings equipment data and engineering simulation into one view.
Build AI factory digital twins that connect engineering and operational data with simulation, using OpenUSD and NVIDIA Omniverse libraries.