DFI QCS051-BCC-4890 Qualcomm QCS6490 2.5″ Pico-ITX SBC with 4GB LPDDR5x Memory and 128GB UFS

Indoor mobile robots, smart cameras and thin clients increasingly need AI inference without the heat and power budget of a separate accelerator. The QCS051-BCC-4890 puts a Qualcomm Dragonwing QCS6490 on a 2.5″ Pico-ITX board from the DFI QCS051 series, rated from 0 to 60 °C for indoor equipment.

The QCS6490 has eight Kryo 670 cores at up to 2.7 GHz and an AI engine rated at 12 TOPS (INT8), which suits object detection or classification on camera feeds running alongside the control software. Memory is 4 GB of soldered LPDDR5x, with 128 GB of UFS 3.1 storage and a microSD 3.0 slot. The Adreno 643L GPU drives one HDMI 1.4 output at 1080p under Linux. Linux Yocto 4.0 is the default image, with Ubuntu 20.04 on request.

Two four-lane MIPI CSI-2 connectors take cameras directly. A GbE port on an ASIX AX88179A controller and two USB 3.1 Gen 1 Type-A ports sit on the edge, with three USB 2.0 ports, an RS-232 port, a CAN FD port and 8-bit DIO on internal headers for drives, sensors and indicator lamps. An M.2 B-key 3052 socket with a nano SIM slot takes 4G or 5G, and an M.2 E-key 2230 socket takes Wi-Fi and Bluetooth. A Type-C download port and a Micro USB debug port serve as service ports for flashing and debugging.

The QCS051-BCC-4890 runs fanless with a heat spreader from 5 V DC on a 2-pin connector, at a typical 3 W and a maximum of 7.25 W, and carries CE, FCC, UKCA and RoHS compliance. For outdoor or unheated equipment, the QCS051-ECC-8890 pairs the same processor with 8 GB of memory and a -25 to 75 °C range. Where 3.5 TOPS is enough, the QCS051-BCC-4830 uses the six-core QCS5430 on the same board.