View source on GitHub

Returns details about a physical devices.

This API takes in a tf.config.PhysicalDevice returned by tf.config.list_physical_devices. It returns a dict with string keys containing various details about the device. Each key is only supported by a subset of devices, so you should not assume the returned dict will have any particular key.

gpu_devices = tf.config.list_physical_devices('GPU')
if gpu_devices:
  details = tf.config.experimental.get_device_details(gpu_devices[0])
  details.get('device_name', 'Unknown GPU')

Currently, details are only returned for GPUs. This function returns an empty dict if passed a non-GPU device.

The returned dict may have the following keys:

  • 'device_name': A human-readable name of the device as a string, e.g. "Titan V". Unlike, this will be the same for multiple devices if each device is the same model. Currently only available for GPUs.
  • 'compute_capability': The compute capability of the device as a tuple of two ints, in the form (major_version, minor_version). Only available for NVIDIA GPUs

device A tf.config.PhysicalDevice returned by tf.config.list_physical_devices or tf.config.get_visible_devices.

A dict with string keys.