DeepX plans Raspberry Pi edge AI card

DeepX, a company based in Korea, is currently working on adapting its M.2 edge AI PCI Express card to be compatible with the popular Raspberry Pi single board computer. This development aims to bring advanced AI capabilities to the Raspberry Pi platform, opening up new possibilities for AI applications in a compact and energy-efficient form factor.

The company has successfully created a 25TOPS edge AI accelerator chip with a remarkably low power consumption of 4.5W. This chip, known as the DX-M1, has already been made available to select early customers for sampling and is specifically designed for vision applications. The DX-M1 chip is housed on a compact 22 x 80mm PCIe3.0 M.2 card, featuring 4GB of low power LPDDR5 memory and 1Gbit NAND Flash accessible via a quad SPI interface.

Looking ahead, DeepX has ambitious plans to develop a cutting-edge 2nm edge AI chip, further pushing the boundaries of AI performance and efficiency. However, one of the key challenges that need to be addressed is the power and cooling requirements of the Raspberry Pi 5 when paired with high-performance AI accelerators like the DX-M1.

With the introduction of the M.2 HAT+ for the Raspberry Pi 5, which supports a PCIe 2.0 interface enabling data transfers of up to 500 MB/s, the ecosystem for AI acceleration on the Raspberry Pi platform is expanding rapidly. This expansion includes offerings from other companies such as Israeli developer Hailo, whose M.2 card utilizes the powerful 26TOPS, 8W Hailo 8 AI accelerator chip.

Despite the competition, DeepX distinguishes itself by showcasing superior thermal performance with the DX-M1 chip operating at a significantly lower temperature compared to the Hailo8 chip under the same workload. Additionally, DeepX provides a comprehensive software development kit that simplifies the process of converting and optimizing Tensorflow or PyTorch models to the ONXX open format for seamless integration with the DX-M1 accelerator chip.

Excitingly, the integration of the DeepX M.2 module with the Raspberry Pi 5 is expected to be officially announced in May, marking a significant milestone in bringing high-performance AI capabilities to the Raspberry Pi community. This collaboration between DeepX and Raspberry Pi is poised to empower developers and enthusiasts to explore new AI applications and innovations in a cost-effective and accessible manner.