Google's Coral division has officially opened pre-orders for the Coral Dev Board Mini, an ultra-compact single-board computer designed for on-device machine learning projects — and a cost-reduced alternative to the original Coral Dev Board.
Announced earlier this year, the Coral Dev Board Mini is designed to reduce footprint and bill-of-materials cost for edge AI projects looking for on-device machine learning acceleration. The compact SBC couples a MediaTek 8167S quad-core Arm Cortex-A35 CPU and Imagination PowerVR GE8300 GPU with Google's own Edge Tensor Processing Unit (TPU) accelerator — offering, the company claims, 4 trillion operations per second (4 TOPS) at 2 TOPS per watt.
The board also includes 8GB of eMMC storage, 2GB of LPDDR3 shared between the system-on-chip and Edge TPU coprocessor, and on-board Wi-Fi 5 and Bluetooth 5.0 connectivity. An on-board microphone is joined by a 3.5mm audio jack, two-pin speaker terminal, micro-HDMI output, MIPI CSI2 four-lane camera port, MIPI DSI four-lane display port, and a 40-pin Raspberry Pi-compatible general-purpose input/output (GPIO) header — plus two USB 2.0 Type-C connectors for power and peripherals.
Its specifications, then, are in places surprisingly higher than the full-size Coral Dev Board: The MediaTek SoC goes toe-to-toe with the NXP i.MX 8M of the original, but the Mini variant offers twice the memory — though 2GB and 4GB variants of the original Dev Board have been promised — and the same Edge TPU coprocessor. The Mini variant's wireless connectivity is stronger, but it lacks the wired gigabit Ethernet port and USB 3.0 connectivity of the full-size version.
The biggest reason to opt for the Dev Board Mini, though, is the price: At $99.99 it's $50 cheaper than the launch price and $30 cheaper than the current price of the full-size Coral Dev Board — but goes up against stiff competition, including the recently-launched NVIDIA Jetson Nano 2GB at just $59. Where the Jetson Nano will win on performance for the majority of workloads, however, the Coral Dev Board Mini wins on footprint and weight.