Simulation goes AI-native: GPU sparse solvers hit GA as digital twins run the furnace

The solver goes GPU

NVIDIA launched cuDSS into general availability on October 21 - a CUDA-X library that moves the solving of large sparse systems, the math-heavy step that repeats throughout engineering and scientific simulations, onto GPUs. It is accessible through C, C++, Python and Fortran interfaces and scales out of the box from individual workstations to large computing clusters, the company says. m:…9223

The launch came with customer benchmarks, all vendor-cited. Applied Materials accelerated sparse solves in Ginestra, its physics-based, defect-centric simulation platform, by up to 10x versus CPU-only approaches. ZEISS SMT achieved a measured 4x speedup in an EUV lithography plasma simulation benchmark using COMSOL Multiphysics with cuDSS on accelerated computing, completing the full simulation in six days and 14 hours. Dassault Systemes reports an 8-10x speedup using cuDSS within its Abaqus structural and electromagnetic solvers. m:…9223

The adopter list is the point: Cadence uses cuDSS in its Celsius software to accelerate simulations of heat, structural stress and warpage in advanced chips and 3D chip designs, helping customers including Nokia; Siemens uses it in Simcenter STAR-CCM+ for fluid-structure interaction workflows; Synopsys uses it in its Sentaurus Device simulator for advanced logic, memory and power chips, helping Infineon model complex devices faster. m:…9223

The twin goes executable

The digital-twin side of the story is about making simulation continuous rather than episodic. Vivix is developing an executable digital twin of its float-glass furnace in collaboration with Siemens, Snowflake and NVIDIA. High-fidelity Siemens Simcenter multiphysics models are converted via reduced-order modeling into lightweight, portable executables deployed natively within Snowflake's AI Data Cloud, while NVIDIA PhysicsNeMo builds AI surrogate models that combine physics-driven causality with simulation and observed data, enabling real-time predictions. m:…6404

Based on a pilot project, Vivix expects to cut furnace energy consumption by 3-5%, extend furnace lifespan by one year, and shorten its data-collection and analytics cycle by 80%. The twin acts as a continuous virtual sensor, reconstructing volumetric heat distribution, gas velocity fields and thermal fluxes across refractory walls in near real time - focused on three connected areas: refractory health, combustion efficiency and glass-melting quality. These are pilot expectations, not measured results. m:…6404

The physics explain the prize. Float-glass furnaces operate at approximately 1,550 to 1,600 degrees Celsius, where teams have limited direct visibility into the interactions among combustion gases, molten glass and the furnace's refractory walls. Deployed on AWS as the underlying cloud provider, the hybrid system makes the unmeasurable reconstructable. m:…6404

Factories get agents

NVIDIA's developer blog added the workflow piece: an agent-assisted approach to building AI-factory digital twins, in which developers bring site plans, equipment assets and engineering requirements into a consistent model that teams can inspect, revise and prepare for simulation. Connecting those inputs into an editable workflow helps teams explore design alternatives and incorporate engineering feedback as the design evolves. m:…1407

The read

Our read: simulation is becoming infrastructure, not a workstation task. GPU-accelerated sparse solvers cut the cost of the inner loop; executable twins cut the cost of the outer loop - continuous sensing and prediction instead of one-off studies. The vendors winning here sell time: up to 10x faster solves, an 80% shorter analytics cycle, and near-real-time visibility into processes that run at 1,600 degrees. m:…9223 m:…6404 m:…1407