Frequently Asked Questions

  1. How do I check that GPU drivers are installed successfully?

    Run import tensorflow and it will show which platform you are running on: Intel® oneAPI Level-Zero (default) or OpenCL™.

    The high level API of TensorFlow tf.config.experimental.list_physical_devices() will tell you the device types that are registered to TensorFlow core.

    $ python
>>> import tensorflow as tf
2021-07-01 06:40:55.510076: I itex/core/devices/gpu/dpcpp_runtime.cc:116] Selected platform: Intel(R) Level-Zero.
>>> tf.config.experimental.list_physical_devices()
[PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU'), PhysicalDevice(name='/physical_device:XPU:0', device_type='XPU')]

  2. How can I see the configurations and rate of utilization of local GPU devices?

    Use the System Monitoring Utility tool to show the capability (clock frequency, EU count, amount of device memory, and so on) of your devices and usage of each sub-module (device memory, GPU engines, and so on).

  3. What’s the relationship of TensorFlow*, Intel® Optimization of TensorFlow* and Intel® Extension for TensorFlow*?

    • TensorFlow is an open-source machine learning library developed and maintained by Google. It is widely used for building and training machine learning models, particularly neural networks.

    • Intel® Optimization of TensorFlow is an optimized library to run TensorFlow on Intel CPUs and replaces stock TensorFlow* for Intel CPUs. Since the TensorFlow 2.9 release, all Intel optimizations for Intel CPUs are upstreamed and available in stock TensorFlow. That means you only need to install stock TensorFlow. DO NOT install both at the same time, the impact is unknown.

      Starting in Q1 2024, the separate Intel® Optimization for TensorFlow* will be discontinued. Intel optimization will be available directly from continuing upstreamed contributions to stock TensorFlow*.

    • Intel® Extension for TensorFlow is an extension of stock TensorFlow* and helps extend acceleration on Intel CPUs and supported Intel GPUs. Intel® Extension for TensorFlow* co-works with stock TensorFlow* (that includes upstreamed optimizations from Intel).

      Currently, Intel® Extension for TensorFlow* has two releases: CPU & XPU.

      • For Intel CPUs, Intel® Extension for TensorFlow* for CPU + stock TensorFlow* provides the best performance of TensorFlow* on Intel CPUs. Install command: pip install --upgrade intel-extension-for-tensorflow[cpu].

      • For Intel GPUs, Intel® Extension for TensorFlow* for XPU + stock TensorFlow* provides the best performance of TensorFlow* on Intel GPUs. Install command: pip install --upgrade intel-extension-for-tensorflow[xpu].

Troubleshooting

This section shows common problems and solutions for compilation and runtime issues you may encounter.

Build from source

Error Solution Comments
external/onednn/src/sycl/level_zero_utils.cpp:33:10: fatal error: 'level_zero/ze_api.h' file not found
#include
^~~~~~~~~~~~~~~~~~~~~		 install level-zero-dev lib level-zero-dev lib is needed when building from source

Runtime

Error Solution Comments
ModuleNotFoundError: No module named 'tensorflow' install TensorFlow Intel® Extension for TensorFlow* depends on TensorFlow
tensorflow.python.framework.errors_impl.NotFoundError: libmkl_sycl.so.2: cannot open shared object file: No such file or directory source /opt/intel/oneapi/setvars.sh set env vars of oneAPI Base Toolkit
version GLIBCXX_3.4.30' not found conda install -c conda-forge gxx_linux-64==12.1.0 install higher version glibcxx