Microsoft makes DirectML acceleration available for TensorFlow
Microsoft has made TensorFlow-DirectML for Windows 10 and Windows Subsystem for Linux widely available. This ends the preview period for the TensorFlow fork, which started in June last year.
With the final release of TensorFlow-DirectML 1.15.5, Microsoft has made available the public package of TensorFlow with a DirectML backend. The package stems from an open source project and to develop the TensorFlow fork, Microsoft collaborated with Intel, AMD and Nvidia. The software enables hardware acceleration for training machine learning models on Windows and the Windows Subsystem for Linux, utilizing the computing power of GPUs that support DirectX 12.
According to Clarke Rahrig, Senior Program Manager at Microsoft’s Windows AI Platform division, the command ‘pip install tensorflow-directml’ in a Python environment is enough to get TensorFlow-DirectML to deploy the GPU. Then the api should work seamlessly with scripts to train models.
| TensorFlow-DirectML on Windows 10 | TensorFlow-DirectML on WSL |
| Windows 10 Version 1709, 64-bit (Build 16299 or later) | Windows 10 Insider Preview, 64-bit (Build 20150 or later) |
| Python x86-64 3.5, 3.6, or 3.7 | Python x86-64 3.5, 3.6, or 3.7 |
|
|
SqueezeNet model sample training in WSL with TensorFlow-DirectML