NVIDIA makes cuDSS generally available for scientific and engineering simulations

NVIDIA has made cuDSS generally available, a CUDA-X library that shifts the solution of large sparse systems to GPUs, a recurring and compute-intensive step in scientific and engineering simulations. The company says the technology can reduce time to results, increase simulation throughput and enable larger, more realistic models at lower cost. The library can be integrated into existing software and scale from workstations to large clusters, with interfaces for C, C++, Python and Fortran.

Applied Materials says it accelerated sparse solves in Ginestra by up to 10x versus CPU-only approaches. In a plasma simulation benchmark for EUV lithography, ZEISS SMT measured a 4x speedup using COMSOL Multiphysics with cuDSS. Dassault Systèmes reports an 8x-10x acceleration in Abaqus solvers; Cadence, Siemens and Synopsys have also integrated the library into their products.