NVIDIA cuDSS (Preview) is an optimized, first-generation GPU-accelerated Direct Sparse Solver library for solving linear systems with very sparse matrices. Direct Sparse Solvers are an important part of numerical computing for real-time applications like autonomous driving and process simulation, where increasing complexity and high throughput demands a robust direct solver.


Key Features

GPU-accelerated Solver

Capitalizing on the CPU’s sequential computing and the GPU’s parallel computing, cuDSS leverages both the CPU and GPU to solve sparse matrices with only a few non-zero elements per row. The result is significantly higher performance than CPU-only solvers .

Core Functionality Support

cuDSS supports single-GPU solving of sparse linear systems, refactorization in cases with multiple systems, as well as different reorderings and types of matrices. cuDSS is also built to be stable, regardless of matrix size. 

Optimized for NVIDIA GPUs

cuDSS supports all NVIDIA GPUs, Pascal and newer, allowing you to integrate direct sparse solvers across a variety of NVIDIA-powered platforms. cuDSS also benefits from the Grace Hopper Superchip architecture.

cuDSS Performance

cuDSS performance benchmark chart

cuDSS is able to achieve significant performance gains compared to CPU-based Direct Sparse Solvers.


Ready to get started with cuDSS?