NVIDIA released cuDSS in general availability on Oct. 21, launching a CUDA-X library designed to accelerate engineering and scientific simulations by shifting sparse linear system calculations from CPUs to GPUs.
The company said the library accelerates the solving of large sparse systems, an intensive mathematical step that repeats throughout simulation workflows for semiconductors, aircraft, and industrial machinery. Offloading these calculations to graphics processors reduces computation times and increases throughput, enabling engineers to run larger models across a wider design space at lower cost.
Several industrial software vendors have incorporated cuDSS into commercial tools, NVIDIA reported. Dassault Systèmes recorded an 8x to 10x speedup when running cuDSS inside its Abaqus structural solver and its computer simulation technology electromagnetic solvers. Applied Materials achieved up to a 10x acceleration over CPU-only setups in Ginestra, its physics-based simulation platform for semiconductor materials and device reliability.
Electronic design automation providers also integrated the library into chip modeling platforms. Synopsys deployed cuDSS in its Sentaurus Device simulator to accelerate modeling of logic, memory, and power chips for Infineon, while Cadence added the library to Celsius to simulate heat, structural stress, and warpage in three-dimensional chip designs for Nokia.
In physics modeling, Siemens added cuDSS to Simcenter STAR-CCM+ to speed up structural simulations in fluid-structure interaction workflows. COMSOL integrated the library into COMSOL Multiphysics, where ZEISS SMT recorded a measured 4x speedup on an extreme ultraviolet lithography plasma benchmark, completing the test in six days and 14 hours.
The library operates out of the box on individual workstations and scales to large computing clusters. Software developers can integrate cuDSS into existing engineering, scientific, and optimization programs using C, C++, Python, and Fortran interfaces.
Live demonstrations of cuDSS are scheduled at the NVIDIA and Dassault Systèmes exhibition booths at GTC Berlin.
