Expanded NVIDIA GPU Support Rounds Out COMSOL Multiphysics Version 6.4

COMSOL has officially released COMSOL Multiphysics version 6.4, which feature improved solver performance through the NVIDIA CUDA direct sparse solver, NVIDIA cuDSS, for NVIDIA AI infrastructure, the new Granular Flow Module, and a new framework for time-explicit dynamic analysis. 

In addition to these major updates, geometry, meshing, and visualization workflows in COMSOL Multiphysics have been enhanced, and productivity has been advanced through optional large language model (LLM)-assisted simulation. The Chatbot window now supports connections to GPT-5, DeepSeek, Google Gemini, Anthropic Claude, and other OpenAI API-compatible models, enabling interactive, model-aware assistance that combines COMSOL documentation with information from the active simulation.

“Our goal with every release of COMSOL Multiphysics is to enhance the user experience and make it easier to achieve faster, more accurate modeling and simulation results,” says Bjorn Sjodin, senior vice president of product management at COMSOL. “With added support for NVIDIA GPU-based solvers, the new Granular Flow Module, and explicit structural dynamics capabilities, we’re looking forward to seeing how users apply the platform to improve product designs and innovate. I’m particularly excited to see how GPU acceleration will speed up compiled simulation apps.”

GPU-Accelerated Solvers

Version 6.4 of COMSOL Multiphysics introduces NVIDIA GPU acceleration for direct solvers and extends multi-GPU capabilities for acoustics simulations. 

cuDSS, a GPU-accelerated sparse direct solver optimized for hybrid CPU–GPU computation, supports all recent NVIDIA GPU architectures. Depending on the hardware and model characteristics, cuDSS can provide speedups compared to CPU-based direct solvers. 

"The integration of cuDSS into COMSOL Multiphysics marks a critical step in bringing accelerated computing to the heart of engineering simulation,” says Tim Costa, general manager of industrial engineering at NVIDIA. “Engineers can now explore larger design space with greater fidelity, reshaping how industries design, validate, and optimize the products that power our world.”

Additionally, NVIDIA CUDA-X cuBLAS library is accelerating the GPU formulation for transient pressure acoustics simulations, which can now be run on multiple GPUs on the same machine, or even on a GPU cluster. 

Simulate Motion of Solid Particles in Bulk Processes

The Granular Flow Module, a new add-on product based on the discrete element method (DEM), enables engineers and researchers to model granular processes such as hopper discharge, silo storage, chute transport, powder spreading, and mixing. By capturing particle-scale effects such as collisions, adhesion, and rotational resistance, and providing detailed control over grain properties, release conditions, and wall interactions, the module helps users evaluate flow uniformity, packing density, mixing efficiency, and wall stresses.

New Capabilities for Time-Explicit Dynamic Analysis

Version 6.4 also introduces a new framework for time-explicit dynamic analysis. The explicit formulation supports a wide range of nonlinear structural materials, including hyperelastic, plastic, viscoplastic, and creep models, and can also be combined with dynamic fracture simulations. 

Additional highlights of COMSOL Multiphysics version 6.4 include: 

• Higher-quality quad-dominant meshing and swept meshing
• Spatially varying transparency
• Array-based plot layouts
• More efficient building of large simulation apps
• New optimization options for time-dependent and parametric studies
• Export of network parameters for deep neural network (DNN) surrogate models
• Surrogate model data generation on clusters
• Import of CFD data in CGNS format
• Frequency- and time-dependent uncertainty quantification

For a closer look at the latest version, click here. Alternatively, for an introduction to COMSOL Multiphysics, check out this blog post. 

Sources: Press materials received from the company and additional information gleaned from the company’s website.