STREAM COMPUTING

Product Features

Instruction sets and processor architectures are specifically customized for deep learning and artificial intelligence applications.
The core computations that accelerate artificial intelligence：convolution and matrix multiplication
High-density computing and parallel storage greatly enhance processing efficiency.
Supporting FP16 vector matrix operations and INT8 quantization operations.
High power efficiency，low latency

Specifications

Computing Ability	FP16 ：128TFLOPS、INT8：256 TOPS
Memory	Memory Type：LPDDR4X Memory Size：16GB Memory Bandwith：119.4GB/S
Interface	PCI Express 4.0 x 16,Support Lane Reversal
TDP	160W
Product Appearance	single-slot full-height（268.44mm x 111.15mm）