Nvidia releases cuDSS for GPU-accelerated engineering simulations

Nvidia made cuDSS generally available on October 21. The CUDA-X software library uses GPUs to accelerate sparse-system calculations that can consume substantial time in engineering and scientific simulations.

Nvidia said Applied Materials achieved acceleration of up to tenfold for sparse solves in its Ginestra simulation platform compared with CPU-only processing. In a separate lithography plasma benchmark, ZEISS SMT used COMSOL software with cuDSS to complete a simulation in six days and 14 hours. Nvidia described that result as a fourfold speed increase.

These calculations recur in simulations used to evaluate designs before manufacturing. Nvidia said accelerating them allows engineers to test more designs and work with larger models. The reported performance gains concern specific applications and benchmarks.

Adoption is already underway across several engineering software suppliers. Nvidia named Cadence, Dassault Systèmes, Siemens and Synopsys among companies integrating cuDSS into tools for chip analysis, structural simulation and other engineering tasks. General availability makes the library available for further integration into simulation software.