CANN - Ascend Community

What Is

CANN__

The AI-oriented, heterogeneous Compute Architecture for Neural Networks (CANN) is a key platform for improving the computing efficiency of Ascend AI processors. It serves as a bridge between upper-layer AI frameworks and lower-layer AI processors and programming. It also offers efficient, intuitive programming APIs for diversified application scenarios, enabling you to quickly build AI applications and services based on the Ascend platform.

Open Architecture

Provides a rich set of APIs, compatible with popular AI frameworks.

Ultimate Performance

Accelerates parallel computing of foundation models and unleashes computing power through hardware-software synergy.

Easy to Use

Provides an intuitive development system based on Ascend C, with unified APIs that are compatible with all hardware series.

CANN Helps Build a Powerful Ascend AI Computing Platform

CANN is at the core of Huawei Ascend AI's basic software and hardware platform, designed to build an all-scenario Ascend AI platform with ultimate performance, intuitive development, and an open ecosystem.

Deep Learning Frameworks

Framework Adapter

Innovative Operators and Domain-Specific Acceleration Libraries

AI Applications

AI Heterogeneous Compute Architecture

AscendCL

Graph Development

Operator Development

Application Development

Compiles and runs computational graphs, providing graph compilation, optimization, loading, and execution capabilities.

Ascend C

Supports simplified operator programming, to suit the habits of operator developers.

AOL

Provides a wide array of high-performance operators with deep optimization and hardware affinity.

HCCL

Provides collective communication solutions for data and model parallelism in single-server multi-device and multi-server multi-device scenarios.

BiSheng Compiler

Provides Host-Device heterogeneous programming and compilation capabilities with Microarchitecure-Aware compilation optimizations.

Runtime

Provides functions such as resource management, media data preprocessing, and model inference, allowing developers to quickly build AI applications.

Driver

MindStudio E2E development toolchain

Ascend AI Processors

Development Scenarios

Graph Development Using AIR

Provides a unified graph development interface for accessing multiple AI frameworks to convert computational graphs into Ascend IR graphs. The use of multiple optimization and acceleration technologies also enables high-performance graph execution of the Ascend AI processors.

Operator Development Using Ascend C

Supports native C and C++ specifications, aligning closely with what developers' are familiar with. Enables cost-effective and high-efficiency operator development, by means of key technologies, such as multi-layer interface abstraction, automatic parallel computing, and twin debugging.

Application Development Using ACL

Provides development APIs for efficient hardware resource management, media data preprocessing, operator calling, and model inference based on C&C++ and Python, helping you easily build high-performance AI applications.