Install the GPU module

This is an optional module that enables a GPU solver specialized for simulating very large systems of granular materials with the penalty-based Discrete Element Method (aka SMC).

Read the introduction to modules for a technical background on the modularity of the Chrono project.

Features

The GPU module allows users to construct a dynamic system consisting of spherical elements

Here are the main features:

  • a variety of friction models
    • frictionless (optimized)
    • single-step pseudo history
    • multi-step history tracking
  • a variety of explicit time integrators
    • forward Euler
    • extended Taylor
    • centered difference
    • Chung
  • single-GPU scaling up to 700 million frictionless elements or 200 million full-history frictional elements
  • trianglular meshes (obj format) in order to facilitate co-simulation with a more full-featured solver (such as the ChSystem)

Requirements

  • To build applications based on this module you must have CUDA installed
  • To run applications based on this module you need
    • a Pascal or newer Nvidia GPU (Pascal and newer are officially supported, though Maxwell should be able to emulate the required features)
    • Linux or Windows
  • This module has been build/tested on the following:
    • Windows 11, MS Visual Studio 17 2022, CUDA 12.3.0 (Ampere and Ada Lovelace GPU architecture)
    • Fedora 40 Linux, GCC 11.3.0, CUDA 12.0.0 (Pascal, Volta, and Ampere GPU architectures)
    • Arch Linux, GCC 12.2.1, CUDA 12.1.0 (Turing GPU architectures)

Building instructions

  1. Repeat the instructions for the full installation, but when you see the CMake window, you must add the following steps:
  2. Set the CH_ENABLE_MODULE_GPU as 'on', then press 'Configure' (to refresh the variable list)
  3. Press 'Configure' again, then 'Generate', and proceed as usual in the installation instructions.

How to use it

  • Look at the API section of this module for documentation about classes and functions.
  • Look at the C++ source of demos to learn how to use the functions of this module.

MacOS support

This module cannot be built for MacOS, Nvidia GPU hardware and CUDA are unsupported.