Hygon Deep Computing Processor DCU (specific model not disclosed)

Official product section /product/accelerator (English block): the Hygon high-performance deep computing processor is based on a mainstream general-purpose parallel computing architecture and can be used with Hygon CPU; it is widely applicable to scientific computing and AI model training and inference. Advantages: greater computing power; open-source ecosystem with a unified underlying hardware driver platform supporting common computing frameworks, libraries and programming models; a hierarchical software stack adapting to different API interfaces and compilers to maximize reuse of mature...

Key Highlights

• Greater computing power: mainstream general-purpose parallel architecture, paired with Hygon CPU
• open-source ecosystem: unified underlying hardware driver platform supporting common frameworks, libraries and programming models
• hierarchical software stack: adapts to different API interfaces and compilers, maximizing reuse of mature AI algorithms and frameworks
• comprehensive software-stack support covering drivers/Runtime, compilers, libraries and mainstream learning frameworks

Technical Specifications

Typical Use Cases

官网未公开 DCU 具体型号与数值规格：/product/accelerator 板块无型号代码、无精度算力(TFLOPS/TOPS)、显存容量/带宽、功耗、PCIe 代数、片间互联带宽等任何数字；PDF 资料需在官网“技术文档下载”提交个人信息表单（姓名/公司/职位/邮箱/手机/验证码）后申请，公开文档列表仅见海光处理器(CPU)产品手册、勘误表及基于海光通用处理器的存储/数据库优化指南，无 DCU 硬件规格 PDF；官网中英双语均为定性板块文案，英文列取自官网英文板块（BLOCK#22），无独立英文产品详情页；图片为官网分层软件栈架构图（官网未提供 DCU 板卡实物/渲染图）；海光 CPU 7000/5000/3000 系列属处理器产品，不属于本采集的四类算力硬件类别(accelerator_card/server/supernode_cluster/devkit_or_module)，未单列条目。