NVIDIA cuDSS Now Generally Available to Accelerate Engineering and Scientific Simulations
NVIDIA cuDSS, now generally available, is a CUDA‑X library that offloads large sparse linear system solves to GPUs, delivering order‑of‑magnitude speedups for engineering and scientific simulations. Reported accelerations include up to 10× on Applied Materials’ Ginestra platform, 4× on a ZEISS EUV lithography plasma benchmark using COMSOL Multiphysics, and 8‑10× within Dassault Systèmes’ Abaqus solvers. The library ships with C, C++, Python, and Fortran APIs and scales from workstations to large GPU clusters, letting developers retrofit existing simulation tools with GPU acceleration. Engineers can thus explore larger, more realistic models faster and at lower cost.
⚡ Key Takeaways
- Applied Materials achieved up to 10× speedup on sparse solves in its Ginestra simulation platform.
- cuDSS provides C, C++, Python, and Fortran interfaces for direct integration into existing codebases.
- Speedups vary (4× to 10×) depending on the application, indicating a trade‑off between workload characteristics and GPU efficiency.
- Integration is done by linking against the cuDSS library and calling its sparse‑solver APIs from supported languages.
- cuDSS scales out‑of‑the‑box from single workstations to multi‑node GPU clusters, but requires GPU hardware and compatible CUDA drivers.
Engineers can retrofit legacy simulation pipelines with cuDSS to cut simulation runtimes by up to tenfold, enabling faster design iteration and broader exploration of design spaces without wholesale rewrites.
✅ Practical Steps
- Link your application against the NVIDIA cuDSS library and replace CPU‑only sparse‑solver calls with cuDSS API calls in C, C++, Python, or Fortran.
- Deploy the updated binary on a GPU‑enabled workstation or cluster to realize the reported 4‑10× speedups on sparse‑solve workloads.
- Validate performance gains on representative benchmarks (e.g., EUV lithography plasma simulation) before scaling to production workloads.
Want the full story? Read the original article.
Read on NVIDIA Blog ↗