On November 28, the 2023 Loongson Product Release and User Conference opened as scheduled at the National Convention Center. With the theme of "Fighting in the Midstream", the conference released the new generation general-purpose processor Loongson 3A6000 and the printer main control chip Loongson 2P0500 on-site, and announced the Loongson processor core IP and Loongson independent command system architecture licensing plan.
More than 4,000 people including Loongson's partners, authoritative media, experts and scholars, and leaders of competent departments gathered at the conference to witness the launch of Loongson's new products and to seek high-level technological self-reliance.
1. Create a mainstream general-purpose CPU chip: Loongson 3A6000 has reached the 10th generation Core quad-core level
According to reports, the Loongson 3A6000 processor adopts the LoongArch autonomous instruction system. It is the first product of the Loongson fourth-generation microarchitecture. The main frequency reaches 2.5GHz. It integrates four newly developed high-performance LA664 processor cores (6 launch dual threads) and supports simultaneous multi-threading technology (SMT2). The whole chip has a total of 8 logical cores.
Integrated security and trustworthy modules can provide secure startup solutions and national secret (SM2, SM3, SM4, etc.) application support.
Hu Weiwu, chairman of Loongson Zhongke, emphasized that the simultaneous multi-threading technology (SMT2) supported by Loongson 3A6000 is a standard technology for mainstream desktop/server CPUs. It allows the CPU core to run multiple threads at the same time, making up for the original shortcomings of Loongson CPUs.
According to the test results of the China Electronics Technology Standardization Institute's Saixi Laboratory, at a frequency of 2.5GHz, the Loongson 3A6000's SPECCPU2006base single-thread fixed/floating point scores reached 43.1/54.6 points respectively, and the multi-process fixed/floating point scores reached 155/140 points respectively;
The SPECCPU2017base single-thread (rate1) fixed/floating point score reaches 5.05/7.78 points respectively, the single-process multi-thread (speed) fixed/floating point score reaches 6.66/18.1 points respectively, and the multi-process (rate8) fixed/floating point score reaches 21.3/21.0 points respectively; the measured bandwidth of Stream exceeds 42GB/s; the measured Unixbench score exceeds 7400 points.
Compared with the previous generation Loongson 3A5000, the single-threaded general-purpose processing performance has been improved by 60%, and the multi-process general-purpose processor performance has been improved by 100%.
Based on relevant test results, the overall performance of the Loongson 3A6000 processor is equivalent to Intel's 10th generation Core quad-core processor launched in 2020.
It should be pointed out that for the CPU, there are two main ways to improve performance.One is to increase the main frequency, and the other is to optimize the core design.
However, due to the current limitations in the development of domestic advanced manufacturing processes, Loongson 3A6000 is still built based on mature processes, and its performance is mainly improved through design optimization.
So we can see that while the performance of the 3A6000 has been greatly improved compared to the previous generation 3A5000, the main frequency remains at 2.5GHz. If Loongson can use domestic advanced process technology in the future, its main frequency will undoubtedly be further improved. At the same time, further optimization of the overlay design is expected to further reduce the performance of CPUs with advanced processes such as Intel and AMD.
Hu Weiwu said: "With the official release of Loongson 3A6000, which has reached the mainstream product level in the market, Loongson has finally completed the 'make-up lesson' on general-purpose processor performance. Loongson 3A6000 has embarked on a path of improving performance through design optimization based on mature technology. The performance of independently developed CPUs can fully catch up with and exceed the level of international mainstream products."
2. The main IP cores of the CPU are independently developed: there is no ceiling for performance!
As a domestic CPU, the Loongson 3A6000 has the highest degree of autonomy and controllability among domestic CPUs.
As early as 2020, Loongson Zhongke launched the self-developed Dragon Architecture (LoongArch) based on twenty years of CPU development and ecological construction, including the basic architecture part and extended parts such as vector instructions, virtualization, and binary translation, with nearly 2,000 instructions.
More importantly, the new Dragon architecture no longer contains the MIPS instruction system. Loongson said that the LoongArch architecture has three characteristics: complete autonomy, advanced technology, and ecological compatibility.
For Loongson, around the self-developed LoongArch instruction set architecture, not only has the self-developed CPU core been launched, but its internally integrated GPU core, encryption and decryption IP, high-speed transmission interface IP, storage interface IP, audio and video interface IP, UART and other interface IP, as well as various specifications of register file, PLL, DDR3/4-PHY, HT-PHY and other hard-core IP are also self-developed.
Zhang Ge, vice president of Loongson Zhongke, further pointed out in an exclusive interview with Core Intelligence after the meeting: "As CPU functions and performance become more and more powerful, it is often not only necessary to solve the problem of the processor core, but also involves a lot of supporting peripheral key IP.
After more than ten years of accumulation, the Loongson teamIt not only has R&D capabilities in instruction sets and CPUIP, but also includes capabilities in 2D/3DGPU, future GPGPU, and AI acceleration.
In addition, communication between the CPU and the outside world also requires a large number of high-speed interface IP, digital-to-analog conversion interface IP, etc. These are all developed by our team ourselves, while most other domestic CPU design manufacturers purchase third-party IP. "
3. Comprehensive coverage of desktop/server/mobile terminals
In addition to the Loongson 3A6000, Loongson Zhongke also announced the server CPU product 3C/D/E6000 and the mobile terminal CPU product 2K3000, which are also based on its fourth-generation "LA664" CPU core.
According to reports, the design of Loongson 3C6000 has been completed. Its single silicon chip has 16 cores and 32 threads (LA664), and its general processing performance has been doubled. The DDR4-3200x4 interface equipped at the same time makes the memory access bandwidth double that of the previous generation 3C5000; the IO performance of PCle4x64 is orders of magnitude higher than that of the previous generation 3C5000. Loongson 3C6000 also supports high-performance national encryption standard encryption and decryption algorithms (SM4 bandwidth >30Gbps).
In addition, in order to improve the interconnection performance between chips, Loongson Zhongke has also launched its self-developed Dragon Link technology (Loongson CoherentLink), which benchmarks current mainstream inter-chip interconnect technologies such as nVlink and CXL. It can achieve higher-speed, lower-latency inter-chip interconnection than I/O buses such as PCIe. This also provides cache coherence protocol transmission for Loongson’s subsequent CPU-CPU interconnection, CPU-GPGPU interconnection, and GPGPU and GPGPU interconnection.
Thanks to the support of Dragon Chain technology, the LS3D6000 dual-silicon 32-core 64-thread and LS3E6000 four-silicon 64-core 128-thread can be quickly realized, while supporting GPU GPU and various accelerator expansions.
In addition, Loongson’s eight-core single-silicon SoC for notebooks/cloud terminals, Loongson 2K3000, has also completed the front-end design. It integrates 8 self-developed LA364 processor cores, with a single-core performance close to 3A5000, and also integrates a self-developed LG200GPGPU core.
According to reports, Loongson’s GPGPU core LG200 can support graphics acceleration, scientific computing acceleration, AI acceleration and other functions. Specifically, it has upgraded the graphics rendering function (OpenGL4.0), supports general computing (supports OpenCL3.0), and supports the INT8 tensor calculation acceleration component. It also has enhanced architecture scalability, with single node performance reaching 256GFlops-1TFlops.
It is worth mentioning that since this year, with the popularity of generative AI, major chip manufacturers such as Intel, Qualcomm, and MediaTek have also promoted generative AI to enter the terminal side, and have launched chips that support the operation of large models of terminal-side generative AI. Intel and Qualcomm are also actively promoting the transition from traditional PCs to AIPCs. Obviously, this is also an opportunity for Loongson Zhongke.
Zhang Ge told Xinzhixun:
"AIPC is a trend, and Loongson will also integrate 8-bit and 16-bit acceleration modules into the next generation of notebook CPUs. We believe that in fact, the chip threshold for this kind of end-side AI itself is not high. The reason why we have not done it now is because this is not the part where we mainly invest our energy. Like the Cambrian team, it turns out that Loongson was once a big team, and their founder was a student of Teacher Hu Weiwu, so in this aspect, we should actually say that it is not difficult to master."
4. 2P0500 printer main control chip
At this conference, Loongson Zhongke also launched a main control SoC chip suitable for single/multi-function printers-Loongson 2P0500.
According to reports, the chip adopts a heterogeneous large and small core structure and integrates DDR3 memory, GMAC, OTG and other functional modules. It has printing data reception, analysis and processing, print engine control, scan timing control, data scanning, image processing, motor control and other functions. A single chip can meet the needs of various typical applications such as printing, scanning and copying.
Loongson Zhongke has launched a variety of solutions for printers, scanners, and copiers based on the Loongson 2P0500, and has cooperated with many domestic mainstream printer manufacturers to complete various application adaptations such as printing, scanning, and copying.
At the conference site, 12 printer manufacturers including Great Wall Information Co., Ltd., CSSC Hanguang Technology Co., Ltd., Shanghai Hantu Technology Co., Ltd., Xi'an University of Electronic Science and Technology, Hengke Technology Industry Co., Ltd., Ningbo Huagao Information Technology Co., Ltd., Yunnan Nantian Electronic Information Industry Co., Ltd., Beijing Chenguang Rongxin Technology Co., Ltd., Beijing Gaodepinchuang Technology Co., Ltd., Tianjin Optoelectronics Communication Technology Co., Ltd., Zhejiang Cangtian Intelligent Information Technology Co., Ltd., and Dalian Zhongying Technology Co., Ltd. signed an agreement with Loongson Zhongke to jointly build a new ecology of domestic printers.
5. A basic software system parallel to X86/Arm has been built
As a LoongArch system that has only been developed for just three years, its software ecosystem is undoubtedly very weak compared to the x86 and Arm ecosystems that have a history of more than 20 years. Therefore, while Loongson actively develops key software (such as browsers) and cooperates with third-party software manufacturers, it also actively embraces the open source software ecosystem to break the situation, and quickly established a complete LoongArch open source ecosystem.
In terms of operating systems, domestic operating system companies such as Tongxin and Kirin have fully supported the new features of Loongson 3A6000 on the basis of continuous compatibility.
In terms of software, Loongson 3A6000 has also improved support for binary translation of software and hardware collaboration, which can improve the binary translation efficiency of the Dragon architecture, run more types of cross-platform applications, and meet various large-scale and complex desktop application scenarios.
Hu Weiwu, chairman of Loongson Zhongke, pointed out in the theme report "Carrying Independence to the End" that the fundamental way out for my country's information industry is to build an independent ecosystem independent of the X86 and Arm systems.
When Hu Weiwu introduced the software ecosystem based on the Dragon architecture of the Loongson independent command system, he believed that the Dragon architecture has built a Linux basic software system alongside X86 and Arm, and has received support from major international software open source communities related to the command system. It has received support from domestic operating systems such as Tongxin, Kirin, Euler, Dragon Lizard, and open source Hongmeng, as well as basic applications such as WPS, WeChat, QQ, DingTalk, and Tencent Conference.
Gao Xiang, vice president of basic software research and development at Loongson Zhongke, said when introducing the open source software work of the Dragon architecture that the Dragon architecture has received widespread support from the international open source software community and has become the world's top instruction set architecture for open source software alongside X86 and ARM.
A large number of important open source software communities, such as the Linux kernel, GCC compilation tool chain, LLVM compiler, Go language, Rust language, QEMU system, V8 JavaScript engine, .NET programming framework, FFmpeg audio and video codec acceleration library, etc., have already implemented support for the Dragon architecture at a higher level and to a more complete degree.
Based on the software versions released by these open source software communities, a Dragon architecture operating system distribution can be directly built.
Loongson Zhongke adheres to the concept of open source ecological construction of openness and cooperation, and has contributed more than one million lines of source code to nearly 200 international open source software project communities. A large number of domestic and foreign developers have also joined the construction of the Dragon Architecture open source ecosystem and made important contributions to the development of the Dragon Architecture version of the open source community. The basic software development of Dragon Architecture has been deeply integrated into the international open source software ecosystem.
Hu Weiwu said: "As the performance of Loongson 3A6000 reaches the level of mainstream products in the market, and the basic software ecosystem based on Dragon architecture is basically completed, Loongson will also embark on a new journey of ecological construction - building an independent information technology system independent of the X86 system and the Arm system."
5. Open authorization of CPU core IP and Dragon architecture instruction system to expand the hardware ecosystem
Hu Weiwu pointed out in the conference report that driven by the policy market, an independent system based on Dragon Architecture has basically been formed, but all links are still relatively weak. A single flower does not mean spring, but a hundred flowers bloom and spring fills the garden. Loongson Zhongke will adhere to the concepts of joint construction, joint consultation and sharing, and build the Dragon Architecture ecosystem with its partners.
To this end, Loongson Zhongke announced that it will open and license the Loongson CPU core IP and the Dragon architecture instruction system to partners to support partners in developing SoC chip products based on the Loongson CPU core IP and the Dragon architecture instruction system.
Specifically,Currently, Loongson CPU core IP has five types, including LA132, LA264, LA364, LA464, and LA664.Wang Wenxiang, chief architect of Loongson Zhongke processor core, said that the performance indicators of these Loongson self-developed series of CPU cores have reached the mainstream level of similar products in the market, and can meet the development needs of SOC chips for information processing, network security, industrial control, edge computing, Internet of Things and other applications.
The ones open to the public for authorization this time are LA132, which is aligned with ArmCortex-M4, LA264, which is aligned with Coretx-A55, and LA364, which is aligned with Coretx-A75.
At this conference, Suzhou Xiongli Technology Co., Ltd., Datang Renewable Energy Experimental Research Institute Co., Ltd., Deyi Microelectronics Co., Ltd., Shandong Linneng Electronic Technology Co., Ltd., Three Gorges Intelligent Control Technology Co., Ltd., National Supercomputing Wuxi Center, Beijing Derui New Technology Co., Ltd., Beijing A total of 10 companies including the Industrial Internet Research Institute of the University of Science and Technology, Xi'an Microelectronics Technology Research Institute, and Northern Automatic Control Technology Research Institute have officially signed cooperation agreements with Loongson Zhongke. They will use Dragon architecture-based CPU cores to design supercomputing chips, special control chips, storage control chips and other SoC chips.
The Dragon architecture software and hardware ecosystem jointly built by Loongson and multiple chip partners is booming, forming a situation where "many trees become a forest".
In addition, Hu Weiwu also revealed at the conference that the Dragon Architecture command system will also consider opening authorization in the future. However, in view of the ecological fragmentation, software incompatibility and other problems caused by the current over-openness of the open source instruction set, Loongson is also drafting a technical specification agreement and publicly soliciting opinions. As long as you sign the technical agreement, you can obtain a permanent license.
At the conference, Asus, the world's leading motherboard brand, also announced that it will combine Asus' rich experience in motherboard design and CPU overclocking to launch motherboard products based on the Loongson 3A6000 chip. At the same time, the person in charge also revealed that the overclocking of Loongson 3A6000 to 3GHz has been verified.
Loongson also joined hands with more than 50 partners to hold a product launch ceremony based on the Loongson 3A6000 processor.
Tongfang Computer, Aerospace 706, Lenovo Kaitian, Beyond Technology, Shengteng Information, Climb, Guoguang Information, Northern Automation, Shirui, Haier Raytheon, Baode Internet Security, Baixin, Huanghe Information Industry, Popular Electronics, Founder Digital, Xiji, Beilian National Core, Aerospace Longmeng, Zhuoyi Hengtong, Yunyong Technology, Shanghai Asus, Shanghai Liulian, JM, High-Power Computer, Tenlink Technology, EMI Storage, Tianan Star Control, Parole, Longmai Technology, Jones Day, Shengbo Technology, Kunshan Jiati, Jiangsu Jiaqing, Jihecheng, Xunwei Electronics, Yuxin Technology, Shenzhen Zhongwei, Hangpu Electronics, Hualong Xunda, Daoli Zhiyuan, Giskeda, Peitian Technology, Intelligent Manifold Robot, Songke Intelligent, Dianke Network Security, Gaohongxinan, Tianrongxin, Amuntai More than 50 partners, including Ke, Kuanyu, Mulian Technology, Quanxunhui, and Changkun Technology, released desktop computers, notebooks, boards, storage products, network security equipment, industrial control computers and other products based on Loongson 3A6000.
6. Comprehensive display, radiating “core” vitality
In the conference exhibition area, nearly 60 Loongson partners exhibited hundreds of Loongson CPU-based solutions, covering multiple scenarios such as information office, industrial control, intelligent manufacturing, smart home, and digital hardware.
In the game experience area, computers equipped with the Loongson 3A6000 processor can support large-scale 3D games such as Cloud Genshin Impact and Tomb Raider. In the office experience area, in addition to common office software such as QQ, WeChat, and DingTalk, industry applications such as ZWCAD, Cloud Desktop, WPS, and digital twin development engine software can also run smoothly on the Loongson computers. All guests at the scene exclaimed, "Loongson computers are easier to use!"
Special experience areas such as the hardware and electronics area, industrial automation control area, and education experience area also brought an "immersive" experience to the guests.
summary:
The performance of the desktop processor Loongson 3A6000 released by Loongson has reached the level of Intel's 10th generation Core quad-core, which also means that this chip will be able to enter a broader mainstream market instead of Loongson's original Xinchuang market. The subsequent server processor Loongson 3C6000 and mobile desktop terminal processor 2K3000 are also expected to enter the mainstream market and compete with Intel and AMD.
Hu Weiwu also said that the Loongson "Three Musketeers" consisting of the desktop processor Loongson 3A6000, the server processor under development Loongson 3C6000 and the mobile desktop terminal processor 2K3000 have certain open market competitiveness.
In addition, Loongson has quickly established a complete LoongArch open source ecosystem around its self-developed Dragon architecture, which is also conducive to the development of Loongson CPUs in the open market. Loongson’s open licensing of self-developed CPU core IP and future Dragon instruction sets will further accelerate the growth of Loongson’s software and hardware ecosystem.
"Loongson CPU is currently the most independent, so there is no risk of 'stuck' or 'ceiling' suppression, and it can be continuously iterated in market practice.
Of course, this also brings some problems, such as building autonomous ecological intelligence on your own and not following suit. But this may also be an advantage for us in the future. I believe that Loongson CPU can transform the advantages of autonomy into performance and ecological advantages, and build a new information technology and software ecosystem that is "three pillars" with the X86 system and the Arm system! Hu Weiwu, chairman of Loongson Zhongke, said with great confidence.