Radxa, Working with Siengine Technology and Arm China, Prepares the Edge AI SiRider S1 SBC

Embedded computing specialist Radxa is preparing to launch a single-board computer designed for machine learning and artificial intelligence (ML and AI) at the edge, in collaboration with Siengine Technology and Arm China: the SiRider S1.

"SiRider is a high-performance industrial single board computer series jointly launched by Radxa and Siengine," Radxa writes of its latest design. "[The SiRider S1] integrates Siengine Technology's industrial-grade '龍鹰一号' [Dragon Eagle One] 7nm AIoT [Artificial Intelligence of Things] application processor SE1000-I, Arm China's self-developed '周易' [Zhou Yi] NPU [Neural Processing Unit], and Radxa's industrial product design experience to provide the market with high-performance and reliable industrial-grade products that can meet the diverse application requirements in the future."

The chunky single-board computer, brought to our attention by Linux Gizmos, features a system-on-chip with a dual-cluster design: the first cluster has four 64-bit Arm Cortex-A76 cores running at up to 2.4GHz plus two lower-power Cortex-A55 cores running at up to 1.8GHz; the second cluster includes two additional Cortex-A55 cores running at the same speed.

Unusually, the chip also includes no fewer than three graphics processing units: an Arm Mali G76-MP10, a lower-performance Mali G76-MP4, and a basic GPU for "2.5D" and two-dimensional work. These operate independently of hardware video codecs, supporting 4k120 decode and 4k60 encode in H.264/H.265 formats. RAM, meanwhile, can be LPDDR4, LPDDR4x, or LPDDR5, with up to 16GB supported; at the time of writing Radxa had not confirmed finalized launch specifications.

Key to the board's position as ideal for artificial intelligence at the edge, though, is the proprietary neural processing unit from Arm China. Dubbed 周易, or Zhou Yi, after the Chinese philosopher behind the I Ching or Book of Changes, the NPU offers two independent accelerator cores rated at 3.34 tera-operations per second (TOPS) and 3.5 TOPS respectively at an unspecified precision — though, in keeping with the rest of the industry, is likely to be measured at INT8.

Elsewhere on the board is a single HDMI output, two gigabit Ethernet ports, four USB 2.0 and four USB 3.0 ports, one four-lane MIPI Display Serial Interface (DSI) and two two-lane Camera Serial Interface (CSI) ports, eMMC and UFS storage options, four lanes of PCI Express Gen. 3, RS232/RS485 serial, and a selection of general-purpose input/output (GPIO) connectivity.

Pricing and availability for the board, the first in a planned series of AIoT single-board computers to come out of the partnership, have yet to be announced; more information is available on the Radxa wiki.