Optimize for Any Hardware with These New Edge Impulse Tools

In our ever-evolving landscape of edge AI, where new processors, development kits and AI co-processor variations emerge rapidly, adaptability is key. We're launching two new methods to help you: "Select AI Hardware" and "Estimate for your custom targets." These will enable our community of users and integrators to continue to innovate at pace.

Configure the latest specs. At the time of writing, the Raspberry PI Pico 2 RP2350 spec is used as the latest released example.

Select AI Hardware

This new feature is inspired by the dynamic evolution of hardware options for edge AI. The process involves defining your target device and application budget, including selecting the target device and processor type and specifying parameters such as clock rate, RAM, ROM, and maximum allowed latency for your application. This configuration is crucial for tailoring your project setup to your hardware's capabilities, thereby effectively estimating your model's performance on the target device.

Custom Targets

Custom Targets allows users to create and customize their target device and application budget according to their project's unique requirements. This involves defining parameters such as the target device type, processor type, clock rate, RAM, ROM, and maximum allowed latency for the application. By tailoring the project setup to the hardware's capabilities, users can ensure optimal performance of their models on the novel target device design.

With Custom Targets, users can unlock the full potential of their hardware, adapt to emerging technologies, and stay at the forefront of edge AI innovation. Whether you're working with the latest development kit or exploring niche coprocessor variations, Custom Targets in Edge Impulse empowers you to harness the power of edge AI like never before.

What are the processor types and architecture options when configuring a custom device?

You can choose from various processor types and architectures when configuring a custom device. Selection determines which options and fields are available to configure your device accurately. For processor types, you have options like Low-end MCU, GPU, AI accelerator, or NPU. The ever-growing processor list of architectures can be specified as well to accommodate such vast ranges as is available within the Arm suite of Cortex-M0+, Cortex-M4F, Cortex-M7, to the Nvidia GPU, AI accelerator, audio-targeted Syntiant NPU, Renesas RA, RISCV like the latest Raspberry Pi Pico 2 RP2350, or spiking neural network-based processors from BrainChip.

How do I get started?

You can change the target device in your project by accessing the configuration form from the top-level navigation and clicking on Change Target device. From there, you can select from a range of processor types, architectures, and clock rates. For example, you could select a Low-end MCU and specify the clock rate, RAM, ROM, and maximum allowed latency for your application.

To configure a custom device in your project, follow these steps:

  1. Access the configuration form from the top-level navigation in your project.
  2. Click on Change Target device.
  3. Select Low-end MCU or another relevant option for a custom device.
  4. Specify the clock rate, RAM, ROM, and maximum allowed latency for your application.

What application budget parameters are configurable?

The application budget parameters you can configure for your custom device are:

Get Started Today

Ready to explore the endless possibilities of edge AI with Custom Targets? Head over to Edge Impulse and discover how you can optimize your impulses for your specific target hardware, ensuring efficient operation and maximizing performance on your device. With Custom Targets, the future of edge AI is in your hands. read on in our documentation in the Select AI Hardware section.

If you have any additions or suggestions while you explore this new feature, please share them using the accompanying thread in our forum.

Comments

Subscribe

Are you interested in bringing machine learning intelligence to your devices? We're happy to help.

Subscribe to our newsletter