3. Quick Start

3.1. Term Interpretation

Term	Description
BM1684	The third-generation tensor processor unit for deep learning developed by SOPHGO
BM1684X	The fourth-generation tensor processor unit for deep learning developed by SOPHGO
TPU	Neural network processing unit in BM1684
SOC Mode	A product form, the SDK runs on A53 AARCH64 platform, and TPU is used as the platform bus device
PCIE Mode	A product form, SDK runs on the host platform ( it can be X86 or AARCH64 server), BM1684 serves as deep learning computing accelerator card in PCIe interface
Driver	Driver is the channel for API to access the hardware
Gmem	DDR memory on card for NPU acceleration
Handle	A user process (thread) handle of the device, and all operations must be done through the handle