The company says cuDSS can be integrated into existing applications through C, C++, Python and Fortran interfaces and can scale from workstations to clusters. NVIDIA cites Applied Materials’ report of up to 10x acceleration in Ginestra versus CPU-only approaches, and a measured 4x speedup in a ZEISS SMT EUV lithography plasma benchmark using COMSOL Multiphysics. These are examples and performance claims presented by NVIDIA; the announcement does not establish that the gains generalize across workloads or configurations. Source: https://blogs.nvidia.com/blog/cudss-generally-available-engineering-scientific-simulation
NVIDIA cuDSS Now Generally Available to Accelerate Engineering and Scientific Simulations
In one minute
NVIDIA announced general availability of cuDSS, a CUDA-X library for solving large sparse systems used in engineering and scientific simulations.
What changed: NVIDIA now says its cuDSS simulation library is generally available, with adoption and reported speedups from engineering-software partners.