DeepX plans Raspberry Pi edge AI card
DeepX in Korea is working on a version of its M.2 edge AI PCI Express card that will work with the Raspberry Pi single board computer.
The company has developed a 25TOPS edge AI accelerator chip with a power consumption of 4.5W that has been sampling to early customers and is aimed at vision applications. The DX-M1 chip is mounted on a 22 x 80mm PCIe3.0 M.2 card with 4GB of low power LPDDR5 memory and 1Gbit NAND Flash via a quad SPI interface.
- DeepX plans 2nm edge AI chip
- Power and cooling challenges of the Raspberry Pi 5
- Moving to a disaggregated architecture
The Raspberry Pi 5 has support for the M.2 format via the M.2 HAT+ with a PCIe 2.0 interface supporting data transfers up to 500 MB/s.
This HAT+ is already used for an M.2 card from Israeli developer Hailo. This uses the 26TOPS, 8W Hailo 8 AI accelerator chip that is fully integrated with Raspberry Pi’s camera software stack and supports out-of-the-box AI applications through Hailo’s software suite and model zoo.
However DeepX claims in tests the other chip operates at a temperature of 60.7 degC compared to 35.5 degC for the DX-M1 for the same YOLO vision AI model. The DeepX software development kit allows Tensorflow or PyTorch models to be converted to the ONXX open format with automatic quantisation and optimisation to a binary file for the accelerator chip to simplify the software chain for developers.
The integration of the DeepX M.2 module with the Raspberry Pi 5 is set to be announced in May.
www.deepx.ai; www.raspberrypi.com
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
