CPU (Software) Rendering#
Open3D’s new visualization functionality (O3DVisualizer
class,
draw()
function and open3d.visualization.gui
and
open3d.visualization.rendering
modules) requires a recent GPU with
support for OpenGL 4.1 or higher. This is not available in certain situations:
GPU is too old to support OpenGL 4.1.
No GPU is available, for example on cloud servers that do not have any GPU (integrated or discrete) installed, or in a docker container that does not have access to the host GPU. This is often the case for many cloud based Jupyter notebooks such as Google Colab, Kaggle, etc.
A GPU is available, but it only supports computation, not graphics. This is a common scenario for cloud based Jupyter notebooks deployed in docker containers.
Open3D supports CPU or software rendering in such situations. Note that this usually produces slower and less responsive rendering, so a GPU is recommended. Currently, this is available only for Linux. There are two separate ways to use CPU rendering depending on whether interactive or headless rendering is desired. Both methods are described below.
Headless CPU Rendering#
For Python code, you can enable CPU rendering for headless rendering when using the :class: .OffscreenRenderer for a process by setting an environment variable before importing Open3D:
- ``EGL_PLATFORM=surfaceless`` for Ubuntu 20.04+ (Mesa v20.2 or newer)
- ``OPEN3D_CPU_RENDERING=true`` for Ubuntu 18.04 (Mesa older than v20.2).
Here are the different ways to do that:
# from the command line (Ubuntu 20.04+)
EGL_PLATFORM=surfaceless python examples/python/visualization/render_to_image.py
# or Ubuntu 18.04
OPEN3D_CPU_RENDERING=true python examples/python/visualization/render_to_image.py
# In Python code
import os
os.environ['EGL_PLATFORM'] = 'surfaceless' # Ubuntu 20.04+
os.environ['OPEN3D_CPU_RENDERING'] = 'true' # Ubuntu 18.04
import open3d as o3d
# In a Jupyter notebook
%env EGL_PLATFORM surfaceless # Ubuntu 20.04+
%env OPEN3D_CPU_RENDERING true # Ubuntu 18.04
import open3d as o3d
Note
Setting the environment variable after importing open3d
will not work,
even if open3d
is re-imported. In this case, if no usable GPU is present, the
Python interpreter or Jupyter kernel will crash when visualization functions are
used.
Note
This method will not work for interactive rendering scripts such
as examples/python/visualization/draw.py
. For interactive rendering see
the next section.
Interactive CPU Rendering#
The method for enabling interactive CPU rendering depends on your system:
You use Mesa drivers v20.2 or higher. This is the case for all Intel GPUs and some AMD and Nvidia GPUs. You should be running a recent Linux OS, such as Ubuntu 20.04. Check your Mesa version from your package manager (e.g. run
dpkg -s libglx-mesa0 | grep Version
in Debian or Ubuntu). In this case, you can switch to CPU rendering by simply setting an environment variable before starting your application. For example, start the Open3D visualizer app in CPU rendering mode with:LIBGL_ALWAYS_SOFTWARE=true Open3D
Or for Python code:
LIBGL_ALWAYS_SOFTWARE=true python examples/python/visualization/draw.py
Note
Mesa drivers must be in use for this method to work; just having
them installed is not sufficient. You can check the drivers in use with the
glxinfo
command.
You use Nvidia or AMD drivers, but your OS comes with recent Mesa drivers (>= v20.2). Install Mesa drivers if they are not installed in your system (e.g. sudo apt install libglx0-mesa in Ubuntu). Preload the Mesa driver library before running any Open3D application requiring CPU rendering. For example:
export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGLX_mesa.so.0 Open3D
Or with Python code:
export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGLX_mesa.so.0 python examples/python/visualization/draw.py
Your OS has old Mesa drivers (< v20.2). We provide the Mesa software rendering library binary for download here. This is automatically downloaded to build/_deps/download_mesa_libgl-src/libGL.so.1.5.0 when you build Open3D from source. The prebuilt version works on Ubuntu 18.04 and Ubuntu 20.04. If you want to use CPU rendering all the time, install this library to
/usr/local/lib
or$HOME/.local/lib
and prepend it to yourLD_LIBRARY_PATH
:export LD_LIBRARY_PATH=$HOME/.local/lib:$LD_LIBRARY_PATH
For occasional use, you can instead launch a program with CPU rendering with:
LD_PRELOAD=$HOME/.local/lib/libGL.so.1.5.0 Open3D
Or with Python code:
LD_PRELOAD=$HOME/.local/lib/libGL.so.1.5.0 python examples/python/visualization/draw.py