Powered by the Sophon BM1684 chip, this computing accelerator card is designed for AI inference. It can be integrated into servers and industrial control computers, efficiently supporting all mainstream AI algorithms on the market to enable applications such as video structuring, facial recognition, behavior analysis, and status monitoring. It delivers AI empowerment to smart city, intelligent transportation, smart energy, smart finance, smart telecom, and smart industrial fields.

Product Specifications

AI Chip

Sophon BM1684

CPU

ARM 8-core A53 @2.3GHz

TPU Architecture

SOPHON

NPU Cores

AI Computing Power

FP32 (FLOPS)

2.2T

INT8 (OPS) Winograd OFF

17.6T

INT8 (OPS) Winograd ON

35.2T

VPU

Video Decoding Capability

H.264: 1080P @960fps

Video Decoding Resolution

H.265: 1080P @1000fps

Video Encoding Capability

CIF/D1/720P/1080P/4K(3840×2160)/8K(8192×4096)/8192×8192

Video Encoding Resolution

H.264: 1080P @70fps
H.265: 1080P @60fps

Video Transcoding Capacity (1080P to CIF)

Up to 18 channels

JPU

JPEG Image Decoding Capability

800 frames/sec @1080P

Maximum Resolution (pixels)

32768×32768

System Interface

Data Link

PCIe X16

Operating Mode

Memory

Specification

LPDDR4

Capacity

12GB

Power Consumption

Typical Power

21W

Maximum Power

30W

Operating Environment

Temperature (Operating/Storage)

-10℃~55℃

Humidity (Operating/Storage)

5~95% non-condensing

Cooling Method

Active cooling

Operating System Support

Ubuntu / CentOS / Debian / UOS / Kylin

Host CPU Support

Compatible with Intel, AMD, Phytium, Hygon, Sunway, Loongson, and other CPUs

Dimensions (with bracket)

169.1mm(W) x 68.9mm(H) x 19mm(D)