NVIDIA cuDSS
NVIDIA's CUDA-X sparse direct solver library, now generally available, with 4-10x reported speedups in engineering simulation.
Overview
NVIDIA cuDSS is a CUDA-X library that moves the solving of large sparse linear systems, a core step in engineering and scientific simulation, onto GPUs. NVIDIA announced general availability on October 21, 2026 at GTC Berlin. Reported results include up to 10x over CPU-only approaches in Applied Materials' Ginestra, 8-10x in Dassault Systemes' Abaqus electromagnetic solvers, and 4x in a COMSOL plasma benchmark for EUV lithography with ZEISS SMT, which finished in six days and 14 hours. It scales from workstations to clusters and is callable from C, C++, Python and Fortran.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- GPU-accelerated solves for large sparse systems
- Integrated with Cadence Celsius, COMSOL Multiphysics, Dassault Systemes Abaqus, Siemens Simcenter STAR-CCM+ and Synopsys Sentaurus Device
- Scales from single workstations to multi-node clusters
- C, C++, Python and Fortran interfaces
Use cases
- Thermal, structural-stress and warpage analysis for advanced chips
- Fluid-structure interaction simulation in STAR-CCM+
- Electromagnetic and plasma simulation in semiconductor process modeling
Pricing
Unknown; the announcement does not state pricing, a version number or supported platforms.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 21 Oct 2026
- Record updated
- 27 Oct 2026
- Record last checked
- 27 Oct 2026