Requirements and Quickstart
Hardware and Hardware Requirements
The minimiums are listed as follows:
Minimium 4GB RAM and 10GB free disk.
Not sensitive to platform. Windows 11 and Ubuntu 22.04 are fully tested.
GPU acceleration is suggested for large dataset computation. Therefore, the requirements for GPU acceleration is listed as follows:
NVIDIA GPUs, with a minimum compute capability of 5.0, i.e., later than GTX 750.
It is suggested that, make
Nx
×Ny
×Nz
×Nt
less than 200,000,000 Lagrangian particles on each Gigabyte of GPU memory for double-side computation under default numerical methods for one still. Generally half it under high-order computations. For example, you can execute on a 600×300×300 mesh for c.a. 80 time steps for a still on a welcomed RTX 2080ti 22G customized GPU, with the price of ~$400.The general short for GPU is VRAM, not computational power, hence old-but-large GPUs are great for use. Newers can be faster, but smallers can run nothing.
The GPU requirements for dynamic LCS are undergoing experiments by the author.
Basic Environment
This page summarizes the system requirements and dependencies of PyFTLE3D
.
Firstly, the following packages are supposed to be installed on your computer manually in advance.
Python version 3.13. Later version than 3.8 till 3.13 should work properly, but was not fully tested. The official download could be found at here. Please select Python installer according to your system framework.
CUDA Toolkit version 12.9 (Download). All versions crossing 12.x.x should theoretically work, but only 12.9 was fully tested. Ealier versions could work as well, but it requires a different version of cupy, and could cause unexpected performance decay and errors. See cupy documentation for more information.
pip newest version, which is used for installing further dependencies. It can be installed by:
python -m ensurepip --upgrade
Generally, the following dependencies can be installed via pip
, the Python package manager.
numpy version 1.21 or later
matplotlib version 3.4 or later
pyvista version 0.32 or later
scipy version 1.7 or later
pyevtk version 1.4 or later
numba version 0.54 or later
tqdm version 4.62 or later
cupy-cuda12x version 13.4.1 or later
pyvistaqt version 0.11.2 or later
PyQt5 version 5.15.11 or later
You can install them after installing pip
by:
pip install -r ./requirements.txt
When it does not work, consider if your current dir is incorrect. It should be run from the project root of Py3DFTLE
when undergoing configurations and under command-line mode. Please feel free about such thing under GUI mode.
Add-on Libs (Optional)
1. ParaView is a powerful beloved open-source visualization platform based on vtk that can be used to visualize the results of 3D FTLE computations. We also integrated the ParaView entrance into our GUI, so you can directly open ParaView from it.
The ParaView installation-free package (v6.0) can be downloaded from official.
For already-installed ParaView, set environment variable PARAVIEW_PATH
pointing to the root dir of ParaView to enable integration from GUI of Py3DFTLE
.
Quickstart
After configurations of environment, you can run a very first demo to test the environment as well as the code, and learn how to use this toolbox for your own data towards new insights.
An computation example coming with a set of our in-house downscaled Direct Navier-Stokes Simulation (DNS) data is provided, including :math:
The dataset can be downloaded from here, and the case descriptions can be referred to paper I.P.
.