GPU Profiler

Usage

Intel® Extension for TensorFlow* provides support for TensorFlow* Profiler, and is enabled by exposing three environment variables (export ZE_ENABLE_TRACING_LAYER=1, export UseCyclesPerSecondTimer=1, export ENABLE_TF_PROFILER=1).

You can use the profiler to track the performance of the TensorFlow* models running on the Intel GPU.

The GPU profiler supports the following profiling items:

• kernel_stats

• tensorflow_stats

• trace_viewer

Example:

• requirements:

  pip install tensorflow_hub
  pip install -U tensorboard-plugin-profile
  

• script:

  import tensorflow as tf
  import numpy as np
  import os
  import tensorflow_hub as tf_hub
  
  logpath = os.path.join('data', 'logs',  'profiler_demo')
  tf.debugging.set_log_device_placement(True)
  
  model = tf_hub.KerasLayer('https://tfhub.dev/google/imagenet/resnet_v1_50/classification/5')
  
  input = tf.ones((1, 224, 224, 3))
  
  with tf.device("/XPU:0"):
      # Warm up
      model(input)
  
      tf.profiler.experimental.start(logpath)
      model(input)
      tf.profiler.experimental.stop()
  
  print('done')
  

• run:

  $ export ZE_ENABLE_TRACING_LAYER=1
  $ export UseCyclesPerSecondTimer=1
  $ export ENABLE_TF_PROFILER=1
  python test.py
  

• Run TensorBoard:
  After executing the above sample code, you will find the log files in ./data/logs/profiler_demo. Then, run TensorBoard with following command:

  tensorboard --logdir=./data/logs/profiler_demo --bind_all
  

• Analyze the result from the Profile tab:

  • kernel_stats:

  • tensorflow_viewer:

  • trace_viewer:

FAQ

1.If you see “No dashboards are activated for the current data set.” the first time you enter the TensorBoard in the browser:

 Refresh the page, and the profile should be shown.