SOPHON Artificial Intelligence Server SA5

Based on Bitmain’s third generation BM1684 cloud AI chip, it’s a high-density, high-performance AI server with great deep learning ability.



Sophon SA5智能服务器可搭载第三方算法,运行视频结构化、人脸检测识别、车辆检测识别等多类安防、视频智能分析应用。SA5智能服务器具备高性能、高计算密度、低功耗等特点,并配备易用的系统软件和工具链,帮助客户进行快速、便捷的二次开发。





Based on the third generation BM1684 chip, much more mature and stable


A single SA5 supports 384~768 channels of video structuring or about 12000 face detection analysis per second

High energy efficiency ratio

SA5 is based on an energy-efficient AI dedicated chip, using a low-power master processor to reduce overall power consumption

High computational density

The single SA5 has Main Processor Unit(MPU),AI Unit(AIU),SATA Storage Unit(SSU), 24 BM1684 chips built in

High expandability

Multiple server clusters can be deployed, and these servers can be connected through Ethernet to enable users to uniformly manage the server, which can be deployed in the cloud

Application Scenarios

City Brain

Safe City

Intelligent Park

City Brain

The city brain includes "AI Center" and "Big Data Center", which provides data, algorithm and hash power for the construction of " City Cerebellum" in various industries, and realizes the auxiliary decision-making, artificial intelligence and data visualization of the entire city brain.

Safe City

The AI analysis platform of video big data will become the focus of safe city construction. The establishment of information chain with face and vehicle information is the core, and video structured server with high performance, high reliability and low power consumption is the key to realize massive video analysis.

Intelligent Park

It uses AI vision technologies such as face and image recognition to realize core management functions, including personnel access, security control, vehicle control, intelligent consumption and intelligent decision-making, and greatly improves the management efficiency of the park with intelligent solutions.


BMNNSDK(BITMAIN Neural Network SDK)一站式工具包,提供底层驱动环境、编译器、推理部署工具等一系列软件工具。易用便捷,涵盖了神经网络推理阶段所需的模型优化、高效运行时支持等能力,为深度学习应用开发和部署提供易用、高效的全栈式解决方案。BMNNSDK,最大程度降低算法和软件的开发周期和成本,使用者能快速在算丰各系列AI硬件产品上部署深度学习算法,便捷实现智能应用。





Peak performance

3*24T Flops@FP32;17.6*24TOPS@int8

Video structuring ability

384~768 channels 1080P

Image analysis ability

12000 images per second

AI Unit(AIU)

Number of intelligent units

Support up to 3

Number of BM1684

Support up to 24

AI computing memory(Total)




Main Processor Unit(MPU)

Master processor

8 cores

Network Interface

2*10Gb Fiber, 2*GbE

Management Interface


Peripheral Interface

2*USB3.0, 1*VGA, RS232

Optional Storage Unit(SSU)

Number of hard disk

2*3.5" HDs(Only support two pieces of AI units when insert SSU)

Chassis Specifications


19-inch 2U

Size of server case (Depth*Width*Height/mm)

438* 619*88.1 (Without handle)

Power supply

AC 220V, Redundant power supply, HotPlugIn, 2*1200W

Typical power consumption