NVIDIA cuDSS Now Generally Available to Accelerate Engineering and Scientific Simulations
NVIDIA Blog Timothy Costa
NVIDIA’s cuDSS is now generally available to speed up engineering and science simulations on GPUs. It cuts the slowest math step, so teams can test more designs faster and at lower cost.
Based on reporting by NVIDIA Blog, Timothy Costa — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
NVIDIA has made cuDSS generally available, turning a key piece of engineering and scientific simulation into a GPU job. The library sits inside CUDA-X and targets the sparse linear solves that show up again and again in large simulation runs. Those are often the bottleneck. When models get more detailed, the wait gets longer, and teams get fewer chances to try new ideas.
The pitch is simple: move that math to accelerated computing and the whole workflow gets more useful. NVIDIA says cuDSS can reduce time to results and raise simulation throughput, which means engineers and researchers can work with bigger, more realistic models and explore a wider design space without pushing costs up as much.
The early users named by NVIDIA are a good sign this isn’t just a lab demo. Applied Materials says cuDSS sped up sparse solves in Ginestra by up to 10x versus CPU-only approaches. COMSOL, working with ZEISS SMT, reported a measured 4x speedup in an EUV lithography plasma simulation benchmark, finishing the full run in six days and 14 hours. Dassault Systèmes says it is seeing 8-10x speedups in Abaqus solvers for performance-critical simulations.
Cadence, Siemens and Synopsys are also using cuDSS in their software, from chip design and heat simulations to fluid-structure interaction and device modeling. NVIDIA says the library can scale from a workstation to a cluster, and developers can hook it into existing tools through C, C++, Python and Fortran. The company is showing it at GTC Berlin, alongside Dassault Systèmes.
My take — AI-written commentary, not fact-checked reporting
This is the unglamorous part of AI-era computing that actually matters: boring math, done faster, on better hardware. The industry loves to talk about flashy models, but the real wins often come from shaving time off the expensive middle of the workflow. GPU acceleration for simulation is not sexy. It is, however, where a lot of the money hides.
Read more about this at: NVIDIA Blog
