From October 2021, the Chinese Academy of Sciences has released the "Special Data Processor (DPU) Technical White Paper" and "Dedic Data Processor (DPU) Performance Datrama Evaluation Method and Implementation" for the entire industry.Why is DPU defined as a disruptive technology for data center development.

In the DPU performance reference evaluation white paper, the evaluation and implementation methods of DPUs in the four areas of the network, storage, computing, and information security are explained. This is also the four major areas of DPU release potential. At the same time, they are just key links in data centers.Therefore, DPU, CPU, and GPUs are called three -driving carriages in the era of compassion economy, and DPUs have become the "third main chip" in the data center field.In addition, intelligent driving, data communication, network security, etc. are also the target application field of DPU.According to the estimation of Sidi.com, the global DPU market will exceed 10 billion US dollars from 2023, with an annual growth rate of over 50%.

Picture source: Sidi.com

This is why Yingweida CEO Huang Renxun focused on the company's first DPU product, "This is just the beginning." At present, not only the industrial giants such as Nvidia are staring at the DPU track, but also startups such as Jingxin Microelectronics also stood out.Show extraordinary technical strength, and the product has been recognized by the industry.

What is DPU?

If you want to know why DPU is hot, you must first figure out, what exactly is DPU?

DPU's English is called Data Processing Unit. The literal translation is the data processing unit, which is also called data center processor by some practitioners.DPU is centered on data. The core role is to accelerate data to help the system solve the low -processing efficiency of CPU processing, and the GPU cannot process loads.

According to the "signal" of the DPU technology white paper, DPU processing data and traditional DSP, baseband processor and other processors, and GPU processing graphics and image data differently, DPU processing data is closer to the underlying architecture, digitalization of various informationEspecially for various timing and structured data, such as large structured forms, data packets in network streams, massive texts, and so on.

DPU architecture reference design

Picture source: DPU technology white paper

DPU potential is huge

It is not difficult to find that the DPU is proposed by solving the bottleneck of data center development as the first outlet. It is the biggest beneficiary of the data center's server to the heterogeneous calculation.With the vigorous development of the data center market, DPU potential in this field is huge.

Taking the Chinese market as an example, the first wave of data centers kicked off in 2014-2015, mainly serving the Internet that was rapidly developing at that time. At that time, many data exchanges appeared in China.From the perspective of specific characteristics, the first wave of data centers is based on calculation. However, a bottleneck appears quickly with the CPU as the main computing power platform.The waste of power resources brings a sense of weakness in computing power.

The second wave of data centers began in 2020, which is what people call the "first year of DPU". The most obvious change is that the data center no longer emphasizes the concept of computing centers and computing power centers.The essence of.In this process, the types of cloud and application algorithms are becoming more and more abundant. Of course, it is still the main goal to improve data processing efficiency.

From "calculation as the center" to "data -centered" is a huge transition, the DPU can stand out from many association processors.As a dedicated engine of data processing, the continuously enriched DPU product type in the future will make the CPU return to the fields that they are good at — overall overall overall and task scheduling, release the computing power of the CPU to the upper application., Algorithm accelerate the sandbox and safety engine.

It can be seen from the figure below that under the data -centered data center framework, the DPU is ubiquitous.

Traditional cloud computing infrastructure and data in the data -based data? The architecture is right?

Picture source: DPU technology white paper

Of course, DPUs need to overcome some challenges in the field of data center.First of all, the types of DPU products at the hardware level need to be rich enough. As described above, the CPU needs to be fully unloaded in the four dimensions of the network, storage, computing, and information security. At the same timeThe overall solution reflects the total cost advantage; at the software level, the current application of the CPU and GPUs in the data center has a standard software framework, which realizes the decoupling of software and hardware. After the DPU is added, this is also a long -term and challenging job.

The above content has mentioned that the terminal application scenario of the DPU is not only in data centers. In the fields of intelligent driving, data communication, network security and other fields, DPUs also have broad application prospects.

For example, in the field of communications, the virtualization of the lord function based on DPU can be applied to 5G edge to calculate UPF, which enhances the high -speed and large bandwidth capabilities of 5G communication in the vertical industry.Realize the virtualization of the collaterals in a fully uninstalled manner, isolated from the host operating system, and the user is exempted from hackers; in the field of autonomous driving, the various virtualization functions provided by DPU can better help developers realize "software softwareDefine cars.

Domestic DPU helps China "new infrastructure"

DPU has opened up a large industrialization trend, and various fields in my country's "new infrastructure" strategy will take the lead in benefiting and accelerate digital transformation.At the same time, "new infrastructure" and "East Digital and West Calculation" will also bring strong demand at the level of data and computing power. In addition, the state's grand layout in 5G and autonomous driving industries, the future development prospects of domestic DPUs are bright.

Grasp the China core opportunity of the DPU industry, Jingxin Micro Electronic Technology (Tianjin) Co., Ltd. (hereinafter referred to as: Jingxinwei) is a young startup, but in terms of DPU technology research and development and product commercial progress, it has reached domestic domestic development.Leading.

Jingxinwei was established in the Tianjin Economic Development Zone in 2020. It focuses on innovative tracks such as new computing, new networks, and new security.Leading DPU products and technical services.

According to the DPU technology white paper, the DPU architecture has three core components, which are control flat, IO subsystems and data planes.In terms of IO subsystem, Jingxinwei provides original technologies such as RAPIDIO embedded system interconnection and software definition interconnection exchange. It has successfully developed RAPIDIO exchange chip NRS1800, software definition interconnection switching chip SDI3210, endogenous security exchange chip ESW5610, Bridge chip PRB0400 and other system IO products.

In the data plane, Jingxinwei has released an integrated AI engine network processing intelligent chip -Jingxin 2820, which mainly faces cloud computing and data center network applications and related tasks.EssenceThe processing of network data packets is the first scenario of DPU, using the "network+computing power" mode as CPU uninstallation. Therefore, DPU is often regarded as an upgraded version of Smartnic (smart network card), but we all know the potential of DPUNever stop there.

Jingxin 2820 realizes the acceleration capabilities of key applications such as cloud computing and the industrial Internet such as cloud computing and industrial Internet.At the same time, the chip has the ability to upgrade the expansion, which can cope with the requirements of large data processing, delay sensitive application requirements, and have rapid deployment capabilities.

Jingxin 2820 supports three major application scenarios:

In this mode of standard network card mode, the chip work in the standard network card method can be configured to be configured to 2*100GE standard network card, 4*10GE network card, 2*25GE or 2*10GE network card, support maximum 16*PCIE4.0, support multi -virtual machine machineSR-IOV data transmission.The software drives the network interface card through the standard Linux network card.

In this mode of intelligent network card mode, the CPU and AI engines participate in data processing, and the chip works in the smart uninstallation mode.Data messages pass through two paths. One is the control path, related protocol packets, and the control message is reposted to the CPU core for the processing of various protocol control faces such as RDMA, OVS, etc.Processing the package/unblocking of the message, accelerated data processing, etc.The AI engine extracted and analyzed input data in real time, testing and identifying whether the data traffic had malicious attacks, and reported and alarm.

Under the mode of independent main control mode, the well -controlled smart chip can work independently, becoming a network main control CPU chip with an AI engine.PCIE works in RC mode, which can connect various peripherals, such as 24 port/48 gigabit exchange chips, SRIO exchange chips, etc.The CPU runs an independent Linux operating system and network protocol stack to realize the main control function of the switch.

The launch of this dual 100G network processing intelligent chip represents a key step in the development of local enterprises in the field of DPU.In the future, Jing Xinwei will continue to pay attention to the three strategic directions of software definition, endogenous security, and brain -like computing, promote the rapid iteration of products such as DPU, and contribute to the country's "new infrastructure" strategy.