NVIDIA cuDSS Now Generally Available to Accelerate Engineering and Scientific Simulations

NVIDIA cuDSS, now available to the public, is a CUDA-X library designed to offload a crucial component of engineering and scientific simulations to GPUs. Complex engineering tasks, ranging from advanced chips to aircraft to industrial systems, rely on simulations to forecast performance before actual construction. However, as these designs grow more intricate, simulations become increasingly time-consuming, straining computational resources and constraining the number of ideas engineers and researchers can explore.

NVIDIA cuDSS addresses this challenge by enhancing the efficiency of solving large sparse systems, a computationally intensive step that recurs throughout these simulations and frequently constrains the scope of testable designs. By leveraging the power of GPUs, cuDSS enables applications to cut down the time to results and boost simulation throughput, empowering users to delve into larger, more realistic models, more precise representations, and a broader design space at a lower cost.

Notable industry leaders have embraced NVIDIA cuDSS to revolutionize their workflows and enhance offerings for customers. Applied Materials has seen up to a tenfold acceleration in sparse solves within Ginestra, its physics-driven, defect-centric simulation platform, using cuDSS, thereby expediting semiconductor materials and device reliability assessments.

Similarly, COMSOL, in collaboration with NVIDIA accelerated computing, observed a fourfold speedup in an EUV lithography plasma simulation benchmark when utilizing cuDSS, completing the entire simulation in six days and 14 hours. Cadence employs cuDSS in its Celsius software to expedite simulations of heat, structural stress, and warpage in advanced chips and 3D chip designs, aiding customers such as Nokia in analyzing complex designs more swiftly.

Dassault Systèmes has reported an 8-10x speedup using cuDSS within its Abaqus structural and electromagnetic simulation technology for performance-critical simulations across industries. Siemens incorporates cuDSS in Simcenter STAR-CCM+ to accelerate fluid-structure interaction simulations. Lastly, Synopsys leverages cuDSS in its Sentaurus Device simulator to accelerate the analysis of advanced logic, memory, and power chips, facilitating Infineon in modeling increasingly complex devices more efficiently. cuDSS can scale seamlessly from individual workstations to extensive computing clusters, allowing developers to incorporate it into existing engineering, scientific, and optimization software.

Integration into these platforms is facilitated through C, C++, Python, and Fortran interfaces. For more information about cuDSS, visit the NVIDIA demo booth and Dassault Systèmes booth at GTC Berlin.