From the source
NVIDIA made its cuDSS library generally available, moving the sparse-solve step of engineering and scientific simulations onto GPUs.
- The CUDA-X library targets large sparse systems, the math-heavy step that repeats throughout simulations and often limits how many designs a team can test.
- NVIDIA says applications can cut time to results and raise simulation throughput, letting users explore larger, more realistic models at reduced cost.
- Applied Materials reports up to 10x faster sparse solves in its Ginestra platform versus CPU-only approaches, and Dassault Systèmes reports an 8-10x speedup in its Abaqus solvers.
- ZEISS SMT measured a 4x speedup in an EUV lithography plasma benchmark run with COMSOL Multiphysics, finishing in six days and 14 hours.
- Cadence, Siemens and Synopsys also use cuDSS in their tools, and developers can integrate it through C, C++, Python and Fortran interfaces, scaling from workstations to clusters.
