NVIDIA cuDSS Reaches General Availability to Speed Up Engineering and Scientific Simulations

GPU-Accelerated Sparse Solvers for Real-World Simulation

NVIDIA has announced the general availability of cuDSS, a CUDA-X library designed to offload sparse linear system solves — a compute-intensive step common in engineering and scientific simulations — onto GPUs. As design complexity grows across semiconductors, aerospace, and industrial systems, simulation runtimes are increasing, limiting how many design iterations teams can explore.

By leveraging GPU acceleration, cuDSS enables significant reductions in time-to-result and higher simulation throughput. This allows engineers and researchers to evaluate larger, more realistic models, improve accuracy, and explore broader design spaces at lower cost.

Industry Adoption Across Key Simulation Platforms

• Applied Materials: Achieved up to 10x speedup in sparse solves within its Ginestra platform for semiconductor reliability analysis.
• COMSOL: ZEISS SMT recorded a 4x speedup in EUV lithography plasma simulations using COMSOL Multiphysics with cuDSS.
• Cadence: Integrated cuDSS into Celsius software to accelerate thermal, stress, and warpage simulations for advanced chips.
• Dassault Systèmes: Reported 8–10x speedup in Abaqus electromagnetic and structural solvers.
• Siemens: Uses cuDSS in Simcenter STAR-CCM+ for fluid-structure interaction workflows.
• Synopsys: Applied cuDSS in Sentaurus Device to speed up simulations for logic, memory, and power devices.

cuDSS scales seamlessly from workstations to large clusters, supporting C, C++, Python, and Fortran interfaces, making it easy to integrate into existing simulation tools.

Why it matters for GPU / AI infrastructure: cuDSS expands the utility of GPU clusters beyond AI training into traditional HPC simulation workloads, increasing ROI for organizations investing in accelerated computing infrastructure.