¶ Overview
Banana Pi BPI-SM9 16-ENC-A3 deep learning computing module is equipped with the highly integrated computing technology processor BM1688 8 Core ARM CA53 1.6GHz, 8GB LPDDR4，32GB eMMC onboard.which has low power consumption, strong computing power, rich interfaces and good compatibility. Support INT4/INT8/FP16/BF16/FP32 mixed precision calculation, can support 16 high-definition video real-time analysis, flexible response to images, speech, natural language and other scenes, can be integrated in intelligent server, edge intelligent computer box, industrial computer, drone, AIOT and other types of products.
¶ SOPHGO BM1688
The SOPHGO processor BM1688 is a highly integrated deep learning vision processor for deep learning reasoning and computer vision. Support 16 TOPS INT8, support mixed precision calculation. It can be integrated in microserver, intelligent computing module, industrial computer, professional network camera, AIOT and other types of products. Efficient adaptation of all deep learning algorithms in the market, image classification, object detection, instance segmentation, semantic segmentation, behavior analysis, text recognition, natural language processing, speech recognition, speech synthesis, search and recommendation applications, for various industries to enable intelligent computing; And integrated image processing hardware: support HDR wide dynamic, 3D noise reduction, 3A, fog removal and other image enhancement, to provide customers with professional video image quality.
Toolchain supports the model format of ONNX/Caffe/TFLite framework; TensorFlow/Pytorch/Paddle/TensorRTBM1684/BM1684X compatible
¶ Key Features
-
High energy efficiency and flexibility
-
16 HD video intelligent analysis
-
Support INT4/INT8/FP16/BF16/FP32 mixed precision calculation
-
-
Super multimedia capability
-
Support 16 HD hard solution, 10 HD hard knitting
-
Supports multiple VI input, audio input, Support ISP
-
Support binocular depth, image stitching, fisheye expansion and other hard acceleration
-
-
Rich peripherals, wide temperature and low power consumption, more flexible deployment
-
support SATA/USB/GbE/HDMI/CAN/SDIO…
-
Lower power consumption, wide temperature design, more adaptable
-
-
Compatible with NV Orin Nano/NX series modules, leading cost performance
-
Pin To Pin compatibility，more plentiful Pinlist/Pinmux
-
4GB/8GB/16GB LPDDR4
-
¶ Getting Started
¶ Hardware
¶ Module hardware interface
¶ BPI-SM9 photo
¶ computing module hardware spec
| Banana Pi BPI-SM9 16-ENC-A3计算模组硬件规格 | |
|---|---|
Module |
SM9 16-ENC-A3 |
CPU |
8 core ARM CA53 1.6GHz |
hashrate |
16 HD video intelligent analysis |
RAM |
8GB LPDDR4x 64bits 4266Mbps |
Flash |
32GB eMMC 5.1 |
Video decode |
H.264&H.265: 16x 1080P@30fps，support 8192 * 8192 / 8K / 4K / 1080P / 720P / D1 / CIF |
Video incode |
H.264&H.265: 10x 1080P@30fps，support 8K / 4K / 1080P / 720P / D1 / CIF |
Picture codec |
JPEG：1080P@480fps，max support 32768 * 32768 |
Audio and video input |
6x VI |
2x I2S 2x Dual channel input and output (optional) |
|
ISP |
2x8Mp30, 2f-HDR, 3DNR / LDC / 3A / Dehaze |
CV |
IVE/VPSS + DPU(Binocular depth) + Stitch(image stitching) + Dewarp |
PCIE |
PCIe Gen3 1 x2 + 1 x2 RC&EP |
USB |
2x USB3.0, 2x USB2.0 |
Network |
2x GbE |
Display |
1x HDMI 2.0 |
Others |
2x SATA3.0, 1x CAN, 1x SD/SDIO, 4x UART, 4x I2C, 6x PWM, 2x SPI, ADC, GPIOs |
Power |
DC 5V |
Operating temperature |
-20℃ ~ +70℃ |
connector |
260-pin SO-DIMM |
size |
69.6mm x 45mm x 6mm |
¶ Core board computing power index
| Banana Pi BPI-SM9 算力指标 | |
|---|---|
System |
Ubuntu20.04 or Linux V4.19 above |
CPU |
BM1688：8 核 ARM A53，1.6GHz |
Data handling capacity |
support FP32/FP16/INT8，16-channel HD video decoding and intelligent analysis can be performed simultaneously |
RAM |
BM1688：8GB，CV186AH：4GB |
LPDDR4， 64bits |
|
data rate 4266Mbps |
|
Flash |
eMMC 32GB |
Video decoding and encoding |
Decode：H.264 & H.265: 16 路 1080P @30fps，maximum resolution8192 * 8192 |
Incode：H.264 & H.265: 12 路 1080P @30fps，maximum resolution 8K |
|
Picture codec |
JPEG：1080P@480fps，maximum resolution 32768*32768 |
¶ Interface definition and comparison with JETSON ORIN NANO
| Banana Pi BPI-SM9 接口定义 | |
|---|---|
JETSON ORIN NANO |
SM9 |
GBE0 |
GBE0 |
GBE1 |
|
RSVD |
|
DP |
|
HDMI |
|
CSI0-1C2D |
CSI0-1C2D |
CSI1-1C2D |
CSI1-1C2D |
CSI2-1C2D |
CSI2-1C2D |
CSI3-1C2D |
CSI3-1C2D |
CSI4-1C2D |
|
CSI5-1C2D |
|
SDMMC |
|
PCIe0-2lane |
|
PCIe0-4lane |
|
PCIe1-2lane |
PCIe1-2lane（2*SATA） |
PCIe2-2lane |
|
USB0-3.0（2.0） |
USB0-3.0（2.0） |
USB1-3.0（2.0） |
USB1-3.0（2.0） |
USB2-3.0（2.0） |
|
UART0 |
UART4 |
UART1 |
UART1 |
UART2 |
UART2 |
SPI0 |
SPI0 |
SPI1 |
SPI1 |
I2C0 |
I2C0 |
I2C1 |
I2C1 |
I2C2 |
I2C2 |
CAM_I2C |
CAM_I2C |
I2S0 |
I2S0 |
I2S1 |
I2S1 |
I2S2（UART0） |
|
CAN |
CAN |
CAM0_MCLK |
CAM0_MCLK |
CAM1_MCLK |
CAM1_MCLK |
CAM2_MCLK |
|
CAM3_MCLK |
|
GPIO*14 |
GPIO*14 |
¶ BPI-SM9 developer’s kit
¶ Development
¶ Source code
-
sophon-demo： https://github.com/sophgo/sophon-demo/tree/release
-
sophon-stream： https://github.com/sophgo/sophon-stream
¶ Resources
-
BM1688 datasheet： https://www.sophgo.com/sophon-u/product/introduce/bm1688.html
-
SOPHGO online course： https://www.sophgo.com/curriculum/online.html
-
SOPHGO on-line case-based learning： https://www.sophgo.com/case-center/index.html