GPU compatibility
GPU acceleration requires the author of a project such as TensorFlow to implement GPU-specific code paths for algorithms that can be executed on the GPU. A GPU-accelerated project will call out to NVIDIA-specific libraries for standard algorithms or use the NVIDIA GPU compiler to compile custom GPU code. Only the algorithms specifically modified by the project author for GPU usage will be accelerated, and the rest of the project will still run on the CPU. For most packages, GPU support is either a compile-time or run-time choice, allowing a variant of the package to be available for CPU-only usage. When GPU support is a compile-time choice, Anaconda will typically need to build two versions of the package, to allow the user to choose between the “regular” version of the project that runs on CPU only and the “GPU-enabled” version of the project that runs on GPU. Due to the different ways that CUDA support is enabled by project authors, there is no universal way to detect GPU support in a package. For many GPU-enabled packages, there is a dependency on thecudatoolkit package. Other packages such as Numba do not have a cudatoolkit dependency, because they can be used without the GPU.
Hardware requirements
NVIDIA released the CUDA API for GPU programming in 2006, and all new NVIDIA GPUs released since that date have been CUDA-capable regardless of market. Although any NVIDIA GPU released in the last 10 years will technically work with Anaconda, these are the best choices for machine learning and specifically model training use cases:- Tesla P100 or V100
- Titan RTX
- GeForce RTX 3050
- Various recent Quadro models
Software requirements
The best performance and user experience for CUDA is on Linux systems. Windows is also supported. No Apple computers have been released with an NVIDIA GPU since 2014, so they generally lack the memory for machine learning applications and only have support for Numba on the GPU.Installing Linux on Windows with WSL
Installing Linux on Windows with WSL
Windows Subsystem for Linux (WSL) lets Windows users install a Linux distribution and run Linux applications, use Linux command-line tools, and more easily interact with Linux-first frameworks, directly on Windows. Several GPU-enabled packages available through Anaconda are either exclusively available for Linux or are primarily optimized for Linux environments. WSL can provide Windows users with a potential way to access these Linux-dependent packages.To run CUDA on WSL and install Linux, follow the steps in NVIDIA’s CUDA on WSL User Guide. You can find additional information about installing and using WSL in Microsoft’s WSL documentation.Once installed, launch WSL, and follow the Linux instructions to install either Anaconda Distribution or Miniconda. Note that this requires a new installation even if you already have Anaconda or Miniconda on Windows, and you’ll need to replicate any environments you want to use in your Linux subsystem.
While WSL can be used to access GPU-enabled packages, this workflow is not officially supported by Anaconda. Users who choose this approach do so at their own risk. For those specifically interested in PyTorch with GPU support on Windows: Anaconda is actively developing a native Windows build of PyTorch with CUDA support, which will be available in an upcoming release.
cudatoolkit conda package in their environment. To select a cudatoolkit version, add a selector such as cudatoolkit=12.1 to the version specification.
Required NVIDIA driver versions, excerpted from the NVIDIA CUDA Toolkit Release Notes:
| CUDA Version | Linux x86_64 Driver Version | Windows x86_64 Driver Version |
|---|---|---|
| CUDA 11.8.x | >= 450.80.02 | >= 452.39 |
| CUDA 12.0.x | >= 525.60.13 | >= 527.41 |
| CUDA 12.1.x | >= 525.60.13 | >= 527.41 |
| CUDA 12.2.x | >= 525.60.13 | >= 527.41 |
Available packages
TensorFlow
TensorFlow is a general machine learning library, but most popular for deep learning applications. There are three supported variants of thetensorflow package in Anaconda, one of which is the NVIDIA GPU version. This is selected by installing the meta-package tensorflow-gpu.
tensorflow package name and will use whatever version of TensorFlow is installed. This makes it straightforward to switch between variants in an environment.
PyTorch
PyTorch is another machine learning library with a deep learning focus. PyTorch detects GPU availability at run-time, so the user does not need to install a different package for GPU support.Caffe (Linux only)
Caffe was one of the first popular deep learning libraries.Chainer/CuPy
Chainer is a deep learning library that uses NumPy or CuPy for computations.XGBoost (Windows/Linux only)
XGBoost is a machine learning library that implements gradient-boosted decision trees. Training several forms of trees is GPU-accelerated.MXNet (Linux only)
MXNet is a machine learning library supported by various industry partners, most notably Amazon. Like TensorFlow, it comes in three variants, with the GPU variant selected by themxnet-gpu meta-package.
Numba
Numba is a general-purpose JIT compiler for Python functions. It provides a way to implement custom GPU algorithms in purely Python syntax when thecudatoolkit package is present.: