Mini-ITX motherboard combines AMD Ryzen Embedded R2314 SoC with two InferX X1 AI accelerators

Flex Logix has just announced the InferX Hawk mini-ITX motherboard equipped with an AMD Ryzen Embedded R2314 SoC and two InferX X1 AI accelerators from the company in order to help customers customize, build and deploy edge and embedded AI systems easily and quickly.

The motherboard features up to 32GB RAM, supports SATA and M.2 MVMe/SATA storage, provides dual Gigabit Ethernet and optional WiFi/4F LTE connectivity, plus the usual USB 3.2, DisplayPort, and serial interfaces, and is meant to compete directly against solutions based on NVIDIA Jetson AGX Xavier.

 

InferX Hawk specifications:

• SoC – AMD Ryzen Embedded R2314 quad-core processor @ 2.1/3.5 GHz (Turbo) with hexa-core Radeon Vega graphics; TDP: 12 to 35W
• AI accelerators – 2x InferX X1 accelerators
• System Memory – 2x DDR4 SO-DIMM up to 32GB capacity
• Storage
  • M.2 M Key socket for NVMe or SATA SSD
  • SATA 3.0 connector
• Video Output – 2x DisplayPort
• Networking
  • Dual Gigabit Ethernet
  • Optional WiFi or 4G LTE via M.2 E-Key socket
• USB – 2x USB 3.1 ports, 2x USB 2.0 ports, 1x USB 3.2 Type-C port
• Serial – 2x COM ports
• Power consumption – 25 to 40W depending on workload and configuration
• Dimensions – 170 x 170 mm (Standard Mini-ITX Form Factor)

On the software side, Flex Logix provides InferX Runtime, and the EasyVision platform running Linux or Windows. The company previously compared the InferX X1 AI accelerator performance to the one of the NVIDIA Jeton Xavier NX, but the Hawk X1 board comprised of two such accelerators is said to outperform Jetson AGX Xavier (configured to run at 30W) in Yolov5 and Resnet50.

The company claims the solution shave development times by over six months compared to a Xavier AGX solution where a custom carrier board and potentially custom chassis may have to be designed, while the Hawk motherboard just needs a standard mini-ITX enclosure. I suppose it’s valid if you can’t use some of the existing carrier boards for Jetson AGX.

The InferX X1 AI accelerator is also available through the InferX X1M M.2 AI accelerator module, so at first, it may feel odd that the company did not try to find a standard mini-ITX board with similar features such as the Kontron D3723-R with Ryzen Embedded R2314, and M.2 sockets using PCIe x4 Gen3/4. But I can see at least two reasons why they have gone the custom board route: it might not have been easy to find boards with two M.2 sockets with PCIe x4 Gen3/4, and M.2 modules have power limitations to about 8W which may limit the performance of the InferX X1 accelerator since a single-chip PCIe card consumes up to 19 Watts.

The InferX Hawk system is designed for a range of computer vision and video applications for safety and security, manufacturing (e.g. optical inspection), traffic and parking management,  logistics healthcare (e.g. medical image analytics), Smart Agriculture, and robotics. Flex Logix did not provide pricing information besides claiming a complete InferX Hawk solution with enclosure would cost $500+ less than a similar system based on the NVIDIA Jetson AGX Xavier module. More details may be found on the product page and press release.

Thanks to TLS for the tip.

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress