|  |  |
|---------------|---------------------------------------------------------------|
| <a href="local_gpu">Local GPU<br/><img class="nav-image" src="images/local-gpu-illustration.png" width=64/></a> | For systems that have a recent, high-end NVIDIA® GPU, TensorFlow is available in a GPU version that takes advantage of the CUDA and cuDNN libraries to accelerate training performance. Note that the GPU version of TensorFlow is currently only supported on Windows and Linux (there is no GPU version available for Mac OS X since NVIDIA GPUs are not commonly available on that platform).  |
| <a href="../../tools/cloudml/getting_started/">CloudML<br/><img class="nav-image" src="images/cloud-ml-illustration.png" width=64/></a> | Google CloudML is a managed service that provides on-demand access to training on GPUs, including the new Tesla P100 GPUs from NVIDIA. CloudML also provides hyperparameter tuning to optimize key attributes of model architectures in order to maximize predictive accuracy. |
| <a href="cloud_server_gpu">Cloud Server<br/><img class="nav-image" src="images/cloud-server-illustration.png" width=64/></a>| Cloud server instances with GPUs are available from services like Amazon EC2 and Google Compute Engine. You can use RStudio Server on these instances, making the development experience nearly identical to working locally. |
| <a href="cloud_desktop_gpu">Cloud Desktop<br/><img class="nav-image" src="images/cloud-desktop-illustration.png" width=64/></a> | Virtual cloud desktops with GPUs are available from Paperspace. This provides an Ubuntu 16.04 desktop environment that you can access entirely within a web browser (note that this requires a reasonably fast internet connection to be usable).|