NVIDIA cuDSS Library Reaches General Availability for Engineering Simulations
The new CUDA-X library accelerates sparse linear algebra solvers on GPUs, helping engineering teams run complex simulations faster and test more design variants.
Key points
- NVIDIA cuDSS is now generally available as a CUDA-X library for shifting simulation workloads to GPUs.
- Early integrators report speedups ranging from four times to ten times compared to CPU-only methods.
- The library supports C, C++, Python, and Fortran interfaces for integration into existing software tools.
NVIDIA announced the general availability of cuDSS, a new CUDA-X library designed to accelerate engineering and scientific simulations by moving critical computational tasks to GPUs. The release aims to help developers and researchers reduce the time required to obtain simulation results while increasing the overall throughput of their modeling workflows.
Complex engineering projects, including the design of advanced microchips, aircraft, and industrial systems, rely heavily on simulation to predict performance before physical prototypes are built. As these designs grow in complexity, the associated simulations take longer to execute, which increases computing demands and restricts the number of design iterations teams can evaluate.
Accelerating Sparse Solvers
NVIDIA cuDSS specifically targets the acceleration of large sparse linear system solves, a math-intensive step that repeats throughout many simulation processes and often becomes a bottleneck. By leveraging GPU power, the library allows applications to process these calculations more quickly. This enables end users to explore larger, more realistic models and a wider range of design options at a lower computational cost.
The library is designed to scale from individual workstations to large computing clusters without requiring significant reconfiguration. Developers can integrate cuDSS into existing engineering, scientific, and optimization software using C, C++, Python, and Fortran interfaces. This approach brings GPU acceleration to tools already familiar to engineers and researchers, reducing the barrier to adopting faster simulation methods.
Industry Adoption and Performance
Several industry leaders have already integrated NVIDIA cuDSS into their software platforms, reporting significant performance improvements. Applied Materials used the library in its Ginestra simulation platform to accelerate sparse solves by up to ten times compared to CPU-only approaches, speeding up the exploration of semiconductor materials and device reliability.
COMSOL Multiphysics users also saw notable gains. ZEISS SMT achieved a four times speedup in an EUV lithography plasma simulation benchmark when using COMSOL with cuDSS on NVIDIA accelerated computing, completing the full simulation in six days and 14 hours. This performance boost helps companies analyze complex physical phenomena more efficiently.
Dassault Systèmes reported an eight to ten times speedup when using cuDSS within its Abaqus solvers for structural and electromagnetic simulations. These performance-critical simulations are used across various industries to ensure designs meet rigorous standards. The integration helps customers analyze complex designs faster, reducing the time between concept and validation.
Other adopters include Cadence, which uses cuDSS in its Celsius software to accelerate simulations of heat, structural stress, and warpage in advanced chips and 3D designs. Siemens integrates the library into Simcenter STAR-CCM+ to speed up fluid-structure interaction workflows. Synopsys uses cuDSS in its Sentaurus Device simulator to help companies like Infineon model increasingly complex logic, memory, and power chips more rapidly.
What should you do?
Engineers and researchers using simulation software should check if their current tools support NVIDIA cuDSS integration. As Threat Times reported on 2026-10-21, NVIDIA has been expanding its infrastructure for data centers and simulations, suggesting a broader push for accelerated computing. Users can visit the NVIDIA demo booth and Dassault Systèmes booth at GTC Berlin to learn more about the library and its applications.
What to do and how to stay safe: NVIDIA
- Verify if your current engineering or scientific simulation software supports integration with NVIDIA cuDSS.
- Review your simulation workloads to identify bottlenecks in sparse linear algebra solves that could benefit from GPU acceleration.
- Consult your software vendor’s documentation for instructions on enabling or configuring cuDSS within your existing tools.
- Monitor NVIDIA’s developer resources for updates on performance benchmarks and best practices for scaling simulations.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is NVIDIA cuDSS?
NVIDIA cuDSS is a CUDA-X library that accelerates the solving of large sparse linear systems in engineering and scientific simulations by utilizing GPU power.
Which programming languages are supported by cuDSS?
Developers can integrate cuDSS using C, C++, Python, and Fortran interfaces, allowing for flexibility in existing software development workflows.
How much faster can simulations run with cuDSS?
Industry partners have reported speedups ranging from four times to ten times compared to CPU-only approaches, depending on the specific application and workload.
