Avnet Unveils the Qualcomm-Powered QCS6490 Vision-AI Development Kit
Powerful yet energy-efficient kit bundles a SMARC 2.1.1-compliant SOM plus compact carrier board for demanding edge AI tasks.
Avnet has announced a new development kit at the Embedded Vision Summit 2024, built around a SMARC-standard system-on-module (SOM) featuring the Qualcomm QCS6490 SoC: the Avnet QCS6490 Vision-AI Development Kit.
Designed, as the name suggests, for projects involving energy-efficient yet high-performance on-device computer vision and artificial intelligence (AI), the Qualcomm QCS6490 at the heart of the board boasts four high-performance Arm Cortex-A78 cores running at up to 2.7GHz alongside four lower-power Cortex-A55 cores running at up to 1.9GHz. There's an Adreno 643 graphics processor running at up to 812MHz, along with a Qualcomm AI Engine neural processing unit (NPU).
It's this latter component that delivers on the development kit's promise of energy-efficient on-device AI: based on Qualcomm's sixth-generation NPU architecture, it delivers 13 tera-operations per second (TOPS) of compute on top of that available from the CPU and GPU. For further acceleration in computer vision tasks there's an Adreno 633 vision processor capable of encoding and decoding video at up 4k30 and 4k60 respectively and a Spectra 570L image signal processor (ISP) capable of running up to five cameras at once.
To this, Avnet has added 8GB of high-speed LPDDR5 memory and 64GB of UFS flash storage on a system-on-module adhering to the SMARC 2.1.1 standard — designed to slot into a bundled carrier board that breaks out two USB 3.1, two USB 2.0, and one USB Type-C On-The-Go (OTG) ports, a gigabit Ethernet port, mini-DisplayPort and MIPI Display Serial Interface (DSI) video outputs, one four-lane and one two-lane MIPI Camera Serial Interface (CSI) inputs alongside two four-lane connectors on the module itself, a four-pin CAN bus header, 40-pin general-purpose input/output (GPIO) header adhering to the Raspberry Pi pinout, and two user-definable buttons and an RGB LED.
For additional storage, the carrier board includes an M.2 M-key slot for Non-Volatile Memory Express (NVMe) boards, while a second M.2 E-key slot allows for the addition of an optional IEEE 802.11ax Wi-Fi 6E and Bluetooth 5.4 radio module. Finally, there's an audio subsystem with two onboard PDM microphones, and an audio codec, plus analog and digital audio connectors. A second gigabit Ethernet port is present on the SMARC SOM, too, but not exposed on the bundled carrier board.
The QCS6490 SMARC and its carrier board are made available as a development kit, which includes a USB type-C Power Delivery (PD) power supply and 12 MP Sony IMX577-based CSI camera module. Optional accessories include additional cameras, Wi-Fi / Bluetooth M.2 modules, NVMe SSDs, and a seven-inch DSI touchscreen display. For software, the kit will ship with a Yocto Linux board support package (BSP) and pre-written AI and multi-camera demos; support for Windows 11 IoT Enterprise has been promised for the second half of the year.
The new development kit is being demonstrated at the Avnet booth (#404) during the Embedded Vision Summit, alongside other live demos including an AI-powered smart parking lot monitor system built with Avnet's RZBoard V2L, edge AI sensor fusion with custom machine learning models from MACSO and Avnet's IoTConnect platform, and a preview of an upcoming development kit based around the AMD Versal Edge AI system-on-chip.
More information on the new QCS6490 Vision-AI Development Kit is available on the Avnet website; pricing had yet to be announced at the time of writing.
Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire: freelance@halfacree.co.uk.