BM1684

Key features

SOPHGO classic TPU processor for AI inference and video analytics in edge servers, smart security and industrial vision
Integrated TPU accelerator supporting multi-precision inference such as INT8, FP16 and FP32
Supports multi-channel video codec and image preprocessing for video structuring and vision analytics systems
Can be integrated through PCIe or board-level interfaces in edge computing devices and servers
Works with SOPHON SDK for model conversion, compilation, runtime inference and multimedia development

Specification	Details
AI	• Integrated SOPHGO TPU acceleration engine• Supports INT8 / FP16 / FP32 inference workloads• Suitable for CNN and classic vision AI models
Video	• Supports multi-channel video decode and encode applications• Video capability depends on card, module and software configuration
Image Processing	• Supports image preprocessing and vision AI pipeline acceleration
Host Interface	• Supports PCIe based accelerator card and embedded product designs
Memory	• Supports external high-bandwidth memory for AI inference workloads• Capacity depends on product design
Software	• Supports SOPHON SDK• Supports BModel compilation, runtime inference and multimedia development