NVIDIA Launches cuDSS Library to Speed Up Engineering Simulations

NVIDIA has announced the general availability of cuDSS, a CUDA-X library designed to shift critical math-heavy steps in engineering and scientific simulations onto GPUs. The release directly impacts software developers and engineering teams handling complex design simulations for advanced chips, aircraft, and industrial systems.

As modern engineering models grow increasingly complex, large sparse systems of equations often become the bottleneck that limits simulation speed and design iteration counts. The cuDSS library specifically targets these sparse solves, allowing host applications to slash calculation times, boost simulation throughput, and lower operational costs while handling larger and more accurate models.

Early Industry Integration and Performance Gains

Several major enterprise software providers have already integrated cuDSS into their simulation pipelines to deliver substantial performance improvements for mutual customers:

• Applied Materials: Accelerated sparse solves in its Ginestra simulation platform by up to 10x compared to CPU-only methods.
• COMSOL & ZEISS SMT: Achieved a 4x measured speedup in an EUV lithography plasma simulation benchmark using COMSOL Multiphysics, finishing the workload in under a week.
• Cadence: Incorporated cuDSS into its Celsius software to speed up thermal, structural stress, and warpage analyses for Nokia and other chip designers.
• Dassault Systèmes: Reported 8x to 10x speedups within Abaqus structural and electromagnetic solvers.
• Siemens: Integrated the library into Simcenter STAR-CCM+ to accelerate fluid-structure interaction workflows.
• Synopsys: Used cuDSS inside the Sentaurus Device simulator to help Infineon model advanced logic, memory, and power chips faster.

Scalability and Developer Access

Designed to scale out of the box, cuDSS supports workloads ranging from single local workstations up to massive enterprise computing clusters. Developers can incorporate the library into their existing engineering, scientific, and optimization software through native C, C++, Python, and Fortran programming interfaces.

Source

Official announcement or documentation