Beijing Baineng Weiye Technology Co., Ltd. (https://bainengweiye.com/) is dedicated to the sales service of international brand original ICs and peripheral components, with over 10 years of experience in electronic component distribution. The company specializes in distributing world-renowned ICs, with products widely used in military and civilian electronic brand fields, and has rich sales experience. The company survives on reputation and conducts business with integrity, ensuring the authenticity of the inventory information published, and all products are brand new in original packaging from manufacturers.
In the current rapid development of artificial intelligence, AI chips, as the core driving force, are becoming the focus of the technology field. Many domestic manufacturers are making efforts and achieving remarkable results in the AI chip field. Today, let us delve into several influential domestic AI chip software manufacturers and understand their AI chip products.
Huawei Ascend: Full-stack Layout, Outstanding Performance
The Huawei Ascend series chips occupy an important position in the domestic AI chip market. Taking the Ascend 910 as an example, it uses its self-developed Da Vinci architecture, specifically designed for artificial intelligence tasks.
In terms of core parameters, the Ascend 910 has a computing power of up to 256T FLOPS (FP16), performing exceptionally well in deep learning training scenarios.
The accompanying graphics card also performs excellently in data processing speed, efficiently cooperating with the chip for large-scale data calculations.
In terms of configuration, Ascend chips are typically combined with Huawei’s self-developed server hardware to create a stable computing platform.
Ascend chips are widely used in data centers, providing powerful computing support for large-scale AI training tasks.
In the field of intelligent security, Ascend chips assist in achieving efficient video analysis and intelligent monitoring; in intelligent transportation, they can achieve accurate traffic flow prediction and intelligent scheduling.
Ascend’s software and hardware stack consists of 5 layers, from the underlying Atlas series hardware and heterogeneous computing architecture to the top layer of AI frameworks, application enablement, and industry applications, comprehensively enhancing performance and intelligence.
In the AI framework layer, it includes the self-developed MindSpore and optimized third-party frameworks (such as PyTorch, TensorFlow, etc.) to achieve efficient operation on Ascend chips.
The application enablement layer includes core components such as ModelZoo, MindX SDK, MindX DL, MindX Edge, and MindIE, facilitating efficient AI development.
Cambricon: AI Chip Pioneer, Technologically Advanced
As a pioneer in the domestic AI chip field, Cambricon’s Siyuan series chips have attracted significant attention. The Siyuan 590 uses a 7nm process technology and possesses powerful computing capabilities.
In terms of core parameters, it can achieve a computing power of 1024 TOPS (INT8), meeting the needs of various complex AI tasks.
In terms of graphics cards, Cambricon chips work in synergy with compatible graphics cards to provide stable support for data processing.
In terms of configuration flexibility, Cambricon chips can adapt to various servers and computing devices, allowing users to build according to their needs.
Cambricon chips are widely used in scenarios such as cloud computing and edge computing. In the cloud, they can be used for large-scale AI model training and inference services, bringing a more convenient intelligent experience.
Cambricon NeuWare is a software development platform specifically designed for its cloud, edge, and endpoint intelligent processor products.
The Neuware software architecture is centered around deep learning applications, relying on mainstream frameworks and integrating Cambricon’s machine learning programming library CNML, Bang programming language, runtime library CNRT, and other multi-layered technologies to provide comprehensive support for multimedia intelligent processing.
At the edge, it can achieve intelligent upgrades for devices such as smart cameras and smart speakers, providing users with enhanced experiences.
Haiguang Information: Compatibility Innovation, Complete Ecosystem
The DCU (Deep Computing Unit) launched by Haiguang Information is unique among domestic AI chips. The Haiguang DCU uses a self-developed DTK (DCU Toolkit) computing platform library, compatible with the open-source ROCm.
This feature allows the Haiguang DCU to possess strong performance while maintaining good software compatibility.
In terms of core parameters, the Haiguang DCU can provide high computing power to meet AI computing needs.
In terms of graphics card configuration, the Haiguang DCU works in synergy with compatible graphics cards to optimize data processing workflows.
The Haiguang DCU is suitable for AI computing tasks in data centers, especially excelling in deep learning inference and training.
Through the Haiguang DTK ecosystem community, developers can access rich resources, including DTK compatible versions, AI ecosystem packages, drivers, and performance analysis toolkits, facilitating the development of applications based on Haiguang DCU.
The Haiguang Light Community, as a leader in the construction of the domestic AI computing ecosystem, provides comprehensive and user-friendly performance and stress testing tools.
Haiguang also offers the Light Source Community, which provides environmental images and model repositories compatible with DCU hardware, ready for download and easy testing.
Tianshu Zhixin: Focused on Computing Power, Breakthrough Innovation
Tianshu Zhixin’s products excel in computing power. For example, one of its chips has high single-card computing power and integration.
In terms of core parameters, it can provide strong computing capabilities to meet various AI computing needs.
In terms of graphics card pairing, Tianshu Zhixin’s chips work with compatible graphics cards to create an efficient computing platform for users.
In terms of configuration, Tianshu Zhixin’s products can flexibly adapt to different servers and computing clusters.
Tianshu Zhixin’s chips are applied in various fields of artificial intelligence, such as in intelligent security, enabling rapid analysis and processing of massive video data; in scientific computing, assisting researchers in complex simulation calculations.
The Tianshu Zhixin IxRT is an inference acceleration engine optimized for mainstream industry frameworks, helping models in vision, speech, recommendation, and natural language achieve excellent inference performance on Tianshu Zhixin cards.
Tianshu Zhixin also provides a GPU virtualization solution based on Kubernetes (K8s) cluster systems, including VGPU solutions.
Suizhi Technology: Software and Hardware Collaboration, Efficient Intelligence
The Suizhi Technology Suisi series chips excel in software and hardware collaboration. The Suisi chips adopt a unique architectural design, demonstrating efficient computing performance in various AI tasks.
In terms of graphics cards and configuration, Suizhi Technology provides a complete solution, ensuring optimal collaboration between chips and surrounding hardware.
Suizhi Technology’s chips are widely used in AI training and inference scenarios in data centers, especially in large-scale deep learning model training, fully leveraging the chip’s potential through software and hardware collaborative architecture design to improve training efficiency.
Suizhi Technology’s computing and programming platform TopsPider, based on innovative programming models, open interfaces, and professional profiling tools, supports adaptive graph optimization strategies, operator generalization, and system-level device virtualization.
At the same time, it has efficient distributed training capabilities, fully tapping into the potential of the Suisi chips through software and hardware collaborative architecture design.
The Enflame GCU hardware-supported software ecosystem matrix includes core heterogeneous acceleration computing platforms (drivers, runtime, compilers, developer tools) as well as extended development libraries, computation graphs, frameworks, and AI development-related toolsets, with rich models, images, solutions, and AI development platforms provided at the upper layer.
These domestic AI chip software manufacturers have made significant progress in technological innovation and product application.
Their AI chip products each have unique characteristics, playing important roles in different application scenarios, providing strong momentum for the development of China’s artificial intelligence industry.
In the future, with continuous technological advancements, it is believed that these manufacturers will continue to launch more competitive products, helping China occupy a more important position in the global AI chip field.
