Development of universal programmable DPU for cloud computing
Focus on the technological development and evolution of data centers. As a general data processor, DPU is not a simple replacement for NIC/SmartNIC, but an integral part of the network infrastructure. Essential change. The DPU itself has the characteristics of universal hierarchical programmability, low-latency network, and unified management and control, so that the DPU is promoting the architectural optimization and reconstruction of the new generation of data centers. As the basic component of general data processing, DPU will offload general data processing tasks originally running in the CPU and GPU, release the computing power of the CPU and GPU, and support the CPU and GPU to achieve greater performance.
##"Cloud Computing Universal Programmable DPU Development White Paper (2023)" The white paper clarifies And analyze the process and current situation of DPU development, and point out which DPU characteristics are the key points to solve the above core problems, thereby promoting the in-depth development of DPU technology and helping to achieve a complete ecological chain construction and industrial implementation.
For DPU technology applications and technical principles, Please refer to the article "DPU technical principles, computing power efficiency and application scenario analysis" and "Mainstream DPU architecture implementation and technology comparison","## DPU Performance Benchmark: Introduction to Evaluation Framework and Test Process (2022)" and "DPU global pattern, The Rise of 5 Domestic Companies (2023)”.
Focuses on analyzing the general programmable features and various application scenarios that DPU needs to have, and also analyzes the limitations of traditional DPU.In recent years, because there is no excellent and mature commercial DPU SoC (System on a chip) solution in the industry, major cloud vendors can only develop their own DPU solutions based on CPU FPGA, resulting in DPU being misunderstood as a fragmented market. , the role and potential of DPU in cloud computing are not correctly understood by the industry. After the "14th Five-Year Plan" clearly pointed out that the construction of new infrastructure should be accelerated, the construction of the Eastern Digital and Western Computing Project and the operator's computing power network came as scheduled. Behind the digital economy, cloud computing is the core computing power base. In cloud computing, DPU has become one of the core components of the infrastructure.
In the era of digital economy, cloud computing continues to enter all walks of life. As the "national team" of cloud computing, China Mobile is increasing its investment to fully support the digital transformation of the government and state-owned enterprises, reduce costs and increase efficiency, and protect the security of state-owned data. In the white paper, Cloud Leopard Intelligent, as the only company invited by China Mobile to participate in the compilation, is the leading domestic DPU chip company and the chip company known in China that can truly achieve high-performance universal programmable DPU SoC. The joint release of this white paper represents the in-depth cooperation between China Mobile and Cloud Leopard Intelligence in the field of DPU, joining forces to contribute to the development of the national cloud infrastructure and DPU.
DPU-centered data center network architectureAs network bandwidth gradually evolves from 25Gbps to 100Gbps, 200Gbps, 400Gbps and even higher bandwidth, traditional data centers have The CPU computing power resources occupied by network data processing are also increasing, and even more than half of them will be consumed in the functions of these infrastructures. Therefore, there is an urgent need for a new type of processor to reduce the consumption of the cloud host CPU. DPU is a general-purpose processor centered on data processing that provides data center infrastructure services. It is the "third main chip" after CPU and GPU. It can offload and accelerate network and storage, and also has basic functions such as security and management. , releasing more computing power resources for customers to use. In cloud computing and data center scenarios, if you need to further improve computing power and maximize infrastructure performance, such as dynamically and flexibly scheduling computing power, network and storage resources, then DPU is necessary and irreplaceable.
At present, most domestic cloud manufacturers still use DPU solutions based on CPU FPGA. These solutions have a certain time advantage in R&D investment, but due to their high power consumption and limited performance, they have not reached the new level. Generation cloud computing requirements. In addition, because FPGA is basically monopolized by the two foreign chip giants, its high price has directly led to high product costs and affected market competitiveness.
DPU SoC products are the ultimate iteration of the former. They require ultra-high heterogeneous chip technology, universal programmability and other features, together with advanced chip technology, to meet more complex, wider and higher-performance applications. need. At present, foreign chip giants and leading cloud service providers have chosen the general DPU SoC product route, because compared to the CPU FPGA solution, DPU SoC has a cost-effective improvement of 4 to 8 times.
Every cloud vendor is looking for the best solution to improve their profits and competitiveness. Because they understand that CPU FPGA is not a long-term solution that can satisfy the new generation of cloud computing, they are all looking forward to a competitive product. Powerful, easy-to-use and cost-effective DPU SoC appears.
The Amazon Cloud (AWS) in the United States not only occupies the highest share of the global cloud computing market, but also was the first to successfully commercialize DPU SoC (AWS calls it Nitro) many years ago. After AWS uses its self-developed DPU SoC, it can earn thousands of dollars more in revenue from selling the computing resources of each server every year. AWS has millions of servers, so the benefits that DPU brings to AWS are huge. The successful application of DPU in AWS has attracted widespread attention in the industry and attracted more and more chip giants to join the DPU track. Nvidia successfully acquired Mellanox, a well-known network chip and equipment company in the industry, for US$6.9 billion in 2020. By integrating Mellanox's network technology, Nvidia quickly launched the BlueField series of DPU SoCs for the global data center market. AMD acquired DPU SoC manufacturer Pensando for US$1.9 billion in 2022. Domestic cloud manufacturers are also seeking technical solutions to evolve from FPGA architecture to general programmable DPU SoC.
It is against this background that China Mobile, China Academy of Information and Communications Technology and Yunbao Intelligence released the "Cloud Computing Universal Programmable DPU Development White Paper (2023)". An in-depth analysis of the development trends of DPU: universal programmability, low-latency network, and unified resource management. At the same time, it also introduces various application scenarios of universal programmable DPU SoC in data centers, operators, heterogeneous computing, etc.
In the construction of domestic data centers, servers are developing from 25G to 100G and higher bandwidth, and the complexity of application deployment continues to increase. It not only needs to support the application management deployment of virtual machines and containers, but also needs to support bare Metal applications. As the core infrastructure component of the data center, DPU must have flexible programming capabilities, high data throughput capabilities, and unified management and control capabilities to meet the current needs of various cloud computing services and data center development.
According to observations from the semiconductor industry, Yunbao Intelligent is currently the chip company known in China that can truly develop high-performance DPU SoC by itself, and its product will also be the first universal programmable DPU SoC chip in China. Not only does it provide data throughput of up to 400G, it is also equipped with a powerful CPU processing unit and works with a variety of programmable data processing engines to achieve hierarchical programmability. According to the description of the white paper, Cloud Leopard Intelligence has mastered and led a number of core technologies in multiple key areas of DPU:
- Programmable high-performance network processing technology
- Programmable low-latency RDMA technology
- DDP (Data Direct Path) data pass-through technology
- Secure computing system
Cloud Leopard Intelligent DPU SoC supports unified operation, maintenance and management of bare metal, virtual machines and containers, provides one-stop solutions such as elastic network and storage, virtualization management and security, greatly improving the cloud The service quality and business flexibility of service providers reduce overall investment and lead the continuous evolution of data centers towards computing and network integration.
China Mobile, as a major cloud service provider supporting the country’s digital economy, gives a clear answer in this white paper. DPU SoC is a key component of cloud computing. The universal programmable DPU SoC can realize the control of data centers. Cost-effective offloading and management of computing, network, and storage resources. It also clearly analyzes the key features that DPU SoC needs to possess: hierarchical programmability, low-latency network, unified management and control, and accelerated offloading to adapt to sustainable development. Cloud vendors are pushing data centers to move towards high efficiency, high scalability, high bandwidth, and high performance. Important technical support for flexibility development. At the same time, it is also the development direction of DPU technology that various cloud vendors are actively researching and exploring.
The above is the detailed content of Development of universal programmable DPU for cloud computing. For more information, please follow other related articles on the PHP Chinese website!

Hugging Face's OlympicCoder-7B: A Powerful Open-Source Code Reasoning Model The race to develop superior code-focused language models is intensifying, and Hugging Face has joined the competition with a formidable contender: OlympicCoder-7B, a product

How many of you have wished AI could do more than just answer questions? I know I have, and as of late, I’m amazed by how it’s transforming. AI chatbots aren’t just about chatting anymore, they’re about creating, researchin

As smart AI begins to be integrated into all levels of enterprise software platforms and applications (we must emphasize that there are both powerful core tools and some less reliable simulation tools), we need a new set of infrastructure capabilities to manage these agents. Camunda, a process orchestration company based in Berlin, Germany, believes it can help smart AI play its due role and align with accurate business goals and rules in the new digital workplace. The company currently offers intelligent orchestration capabilities designed to help organizations model, deploy and manage AI agents. From a practical software engineering perspective, what does this mean? The integration of certainty and non-deterministic processes The company said the key is to allow users (usually data scientists, software)

Attending Google Cloud Next '25, I was keen to see how Google would distinguish its AI offerings. Recent announcements regarding Agentspace (discussed here) and the Customer Experience Suite (discussed here) were promising, emphasizing business valu

Selecting the Optimal Multilingual Embedding Model for Your Retrieval Augmented Generation (RAG) System In today's interconnected world, building effective multilingual AI systems is paramount. Robust multilingual embedding models are crucial for Re

Tesla's Austin Robotaxi Launch: A Closer Look at Musk's Claims Elon Musk recently announced Tesla's upcoming robotaxi launch in Austin, Texas, initially deploying a small fleet of 10-20 vehicles for safety reasons, with plans for rapid expansion. H

The way artificial intelligence is applied may be unexpected. Initially, many of us might think it was mainly used for creative and technical tasks, such as writing code and creating content. However, a recent survey reported by Harvard Business Review shows that this is not the case. Most users seek artificial intelligence not just for work, but for support, organization, and even friendship! The report said that the first of AI application cases is treatment and companionship. This shows that its 24/7 availability and the ability to provide anonymous, honest advice and feedback are of great value. On the other hand, marketing tasks (such as writing a blog, creating social media posts, or advertising copy) rank much lower on the popular use list. Why is this? Let's see the results of the research and how it continues to be

The rise of AI agents is transforming the business landscape. Compared to the cloud revolution, the impact of AI agents is predicted to be exponentially greater, promising to revolutionize knowledge work. The ability to simulate human decision-maki


Hot AI Tools

Undresser.AI Undress
AI-powered app for creating realistic nude photos

AI Clothes Remover
Online AI tool for removing clothes from photos.

Undress AI Tool
Undress images for free

Clothoff.io
AI clothes remover

Video Face Swap
Swap faces in any video effortlessly with our completely free AI face swap tool!

Hot Article

Hot Tools

VSCode Windows 64-bit Download
A free and powerful IDE editor launched by Microsoft

MinGW - Minimalist GNU for Windows
This project is in the process of being migrated to osdn.net/projects/mingw, you can continue to follow us there. MinGW: A native Windows port of the GNU Compiler Collection (GCC), freely distributable import libraries and header files for building native Windows applications; includes extensions to the MSVC runtime to support C99 functionality. All MinGW software can run on 64-bit Windows platforms.

mPDF
mPDF is a PHP library that can generate PDF files from UTF-8 encoded HTML. The original author, Ian Back, wrote mPDF to output PDF files "on the fly" from his website and handle different languages. It is slower than original scripts like HTML2FPDF and produces larger files when using Unicode fonts, but supports CSS styles etc. and has a lot of enhancements. Supports almost all languages, including RTL (Arabic and Hebrew) and CJK (Chinese, Japanese and Korean). Supports nested block-level elements (such as P, DIV),

PhpStorm Mac version
The latest (2018.2.1) professional PHP integrated development tool

SublimeText3 English version
Recommended: Win version, supports code prompts!