Specifications
Processor
SOPHGO BM1688 deep-learning processor, 8-core Arm Cortex-A53 @ 1.6 GHz
AI Performance
16 TOPS (INT8), 4 TFLOPS (FP16)
Precision Support
INT4, INT8, INT16, FP16, BF16, FP32 mixed precision
Memory
8GB LPDDR4x, 64-bit @ 4266 Mbps
Storage
32GB eMMC 5.1 plus 1x TF card interface
Video Decode
H.264/H.265 up to 16x 1080p @ 30 fps; resolutions to 8192 x 8192 (8K/4K/1080P/720P/D1/CIF)
Video Encode
H.264/H.265 up to 10x 1080p @ 30 fps; supports 8K/4K/1080P/720P/D1/CIF
JPEG Acceleration
1080p at up to 480 frames per second; image sizes to 32,768 x 32,768 pixels
On-chip ISP
Dual 8 MP inputs with 2-frame HDR, 3D noise reduction, lens distortion correction, dehazing and 3A
Camera Input
6x VI inputs; 2x I2S (dual-channel input/output, optional)
PCIe
PCIe Gen3 — 1x2 + 1x2 lanes, root complex and endpoint modes
Networking
2x Gigabit Ethernet
Display
1x HDMI 2.0
I/O Interfaces
2x SATA 3.0, 1x CAN, 1x SD/SDIO, 4x UART, 4x I2C, 2x SPI, 6x PWM, ADC, GPIO
USB
2x USB 3.0, 2x USB 2.0
Module Connector
260-pin SO-DIMM, pin-compatible with NVIDIA Jetson Orin Nano / Orin NX carriers
Power Input
12 V / 3 A DC
Operating Temperature
-20 °C to +70 °C
Dimensions
69.6 mm x 45 mm x 6 mm
Toolchain
ONNX, Caffe and TFLite model formats
Frameworks
TensorFlow, PyTorch, Paddle, TensorRT, BM1684 and BM1684X development environments
Target Applications
Microservers, edge systems, industrial platforms, AIoT devices and drone payloads
Overview
The Banana Pi BPI-SM9 16-ENC-A3 is Banana Pi's first compute module built on the SOPHGO BM1688 deep-learning processor, designed for multi-channel video analytics where a dedicated inference accelerator is too small and an x86 box PC is too power-hungry. The module pairs eight Arm Cortex-A53 cores with a 16 TOPS INT8 engine supporting INT4 through FP32 mixed precision.
Its defining feature is the 260-pin SO-DIMM interface, which Banana Pi documents as pin-compatible with NVIDIA Jetson Orin Nano and Orin NX carrier boards while exposing a wider interface set than the Jetson equivalents — notably dual Gigabit Ethernet, dual SATA 3.0 and a six-input video front end. Integrators with existing Orin carriers can evaluate the BM1688 platform with a board swap rather than a redesign.
Video throughput is the headline for surveillance and transportation workloads: 16 simultaneous 1080p30 decodes, 10 simultaneous 1080p30 encodes, an integrated dual-8MP ISP with HDR, 3DNR, lens correction and dehazing, and JPEG acceleration at 480 frames per second. This allows a single module to serve as an NVR-class analytics engine.
Software support covers ONNX, Caffe and TFLite model import with a toolchain compatible with TensorFlow, PyTorch, Paddle, TensorRT and the earlier BM1684/BM1684X generations, easing migration from deployments already running SOPHGO silicon. Contact QS Compute for module pricing, carrier board availability and development kit bundles.
Key Benefits
Jetson Orin pin-compatible 260-pin SO-DIMM module for fast platform evaluation; 16-channel 1080p30 decode plus dual SATA and dual GbE not available on comparable Jetson carriers; -20 °C to +70 °C industrial rating; a mature BM1684/BM1684X-compatible toolchain that reduces porting effort; low-power 12 V / 3 A operation suited to fanless enclosures.
Applications
Multi-channel video analytics and NVR-class edge servers; intelligent transportation and ANPR; industrial machine vision; AIoT gateways and microservers; drone and unmanned platform payloads.
Request a Quote — BANANA PI BPI-SM9 16-ENC-A3 — SOPHGO BM1688 EDGE AI COMPUTE MODULE
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →