Device Management
=================

**eIQ AI Hub** provides two types of device access:

- **Public devices**: Maintained and hosted by NXP, available to everyone for immediate use without registration

- **Private devices**: Exclusive devices for private users. Private devices allow you to perform deep optimization and testing with dedicated access permissions.

Key Features
------------

* Private devices support **ML Runtime Software OTA** - update the ML runtime environment on private devices without manual intervention
* Benchmark on private devices to get real-world performance data

Supported Devices
-----------------

The eIQ AI Hub supports a wide range of NXP platforms:

.. list-table::
   :header-rows: 1
   :align: left
   :widths: 15 28 15 25 8

   * - Platform
     - CPU
     - Memory
     - NPU
     - eTOPS
   * - i.MX 95
     - 6x Cortex-A55 + Cortex-M7 + Cortex-M33
     - 16GB LPDDR5
     - Neutron
     - 8
   * - i.MX 943
     - 4x Cortex-A55 + 2x Cortex-M7 + 2x Cortex-M33
     - 8GB LPDDR5
     - Neutron
     - 2
   * - i.MX 93
     - 2x Cortex-A55 + Cortex-M33
     - 2GB LPDDR4
     - Ethos-U65
     - 1.8
   * - i.MX 8M Plus
     - 4x Cortex-A53 + Cortex-M7
     - 4GB LPDDR4
     - VSI NPU
     - 2.3
   * - i.MX 8M Plus RDM
     - 4x Cortex-A53 + Cortex-M7
     - 4GB LPDDR4
     - Ara240 + VSI NPU
     - 40 + 2.3
   * - i.MX RT700
     - Cortex-M33 + HiFi 4 DSP
     - 7.5 MB SRAM
     - Neutron
     - 0.17

Public Devices
--------------

.. image:: /_static/images/device_management/public-device.gif
   :alt: Public Device Usage Demo
   :width: 100%

Public devices are shared resources maintained by NXP. They allow you to:

* Quickly test and validate models on real NXP hardware
* Profile model performance without owning physical devices
* Access the latest NXP platforms and accelerators

Simply select a public device when creating a profiling or benchmark job to get started immediately.

Private Devices
---------------

Private devices provide dedicated access for in-depth development and testing. To use private devices, you need to register your device first.

Device Registration
~~~~~~~~~~~~~~~~~~~

.. image:: /_static/images/device_management/device-registration.gif
   :alt: Device Registration Demo
   :width: 100%

To register a private device:

1. Navigate to the Device Management section in eIQ AI Hub
2. Click "Add Device" and follow the registration wizard
3. Download the device registration script
4. Run the script on your device to complete registration
5. Your device will appear in the private devices list once registered

ML Runtime Software OTA
~~~~~~~~~~~~~~~~~~~~~~~

.. image:: /_static/images/device_management/ml-runtime-ota.gif
   :alt: ML Runtime OTA Demo
   :width: 100%

Private devices support Over-The-Air (OTA) updates for ML runtime software. This feature allows you to:

* Update the ML runtime environment remotely
* Install new versions of inference engines
* Apply patches and optimizations without physical access

To trigger an OTA update:

1. Select your private device from the device list
2. Navigate to the "ML Runtime OTA" tab
3. Choose the target ML runtime version
4. Click "Install Update" to start the OTA process

Benchmark on Private Devices
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. image:: /_static/images/device_management/benchmark-private-device.gif
   :alt: Benchmark on Private Device Demo
   :width: 100%

Benchmarking on private devices provides accurate, real-world performance metrics:

* Run benchmarks in your actual deployment environment
* Test with custom configurations and settings
* Compare performance across different optimization strategies
* Validate results with production-like workloads

To run a benchmark on your private device:

1. Upload and optimize your model
2. Select your registered device from "My Devices" as the target platform
3. Configure the benchmark parameters
4. Click "Submit Task" to start the benchmark
5. Navigate to the task details to view the results

.. note::

   For a complete overview of benchmarking, see :doc:`Benchmark the model<./benchmark>`

Next Steps
----------

* :doc:`Profile the model on real NXP hardware<./ondevice_profiling>`
* :doc:`Optimize your model<./optimize>`
* :doc:`Upload your model<./upload_model>`
* :doc:`Upload your dataset<./upload_dataset>`