NVIDIA says its upcoming Vera Rubin HPC platform will accelerate the simulations used by scientists and engineers to evaluate candidate materials, molecules and designs. The system combines more than 50,000 CUDA cores in a single GPU with sparse-data acceleration and seventh-generation NVIDIA NVLink interconnect technology. Rubin HPC is expected to be available in 2028.
As agents help researchers explore more possibilities, simulations can test ideas and guide the next steps. NVIDIA says Rubin HPC is designed to shorten the path from exploration to scientific discovery and industrial design by speeding up those simulations.
Rubin HPC GPUs feature a record number of CUDA cores—more than 50,000 on a single GPU. Sparse-data acceleration is intended to keep those cores supplied with data, while seventh-generation NVIDIA NVLink technology accelerates GPU-to-GPU connections.
“Rubin HPC accelerates the simulations that agents use to test ideas and guide next steps, helping scientists and engineers explore more possibilities and bring breakthroughs closer,” said Ian Buck, vice president of hyperscale and high-performance computing at NVIDIA.
Turning faster computation into faster discovery
Scientific computing workloads in areas such as climate modeling, engineering simulation, genomic analysis and financial risk often follow a similar pattern. An application accesses a very large dataset, collects a small number of scattered values, performs a calculation and repeats the process at large scale.
Rubin HPC is designed to address this challenge by scaling data movement and computation together, rather than increasing compute capacity alone.
This balance is intended to optimize Rubin HPC for scientific computing. It helps keep the GPU’s large number of CUDA cores supplied with data, allowing them to spend more time computing instead of waiting. NVIDIA says this can accelerate repetitive patterns found in simulation workloads, including sparse matrix operations, stencil computations, sorting and reduction.
These patterns are common across data analytics, scientific and engineering workloads. NVIDIA says dedicated GPUs with broad acceleration capabilities can help support breakthroughs across industries.
NVIDIA CUDA-X provides optimized implementations of complex algorithms used by agents, including algorithms relevant to simulation and analysis. Used with Rubin HPC, the libraries are intended to help agents reach reasoning outcomes more quickly.
Faster agent-assisted simulations could also provide quicker feedback, shortening the hypothesis–simulation–refinement cycle. Researchers and engineers would then be able to select their next steps sooner and focus laboratory work on the most promising candidates.
Expected availability
Rubin HPC is expected to be available in 2028.
Source: blogs.nvidia.com


