Design scheme of Loongson chip processor

The domestic operating system UOS was officially released not long ago, while the domestic new generation processor was officially released in December 2019. Recently, the official document introduced its product features and new features.
As one of the representatives of “China Core”, Loongson released a new generation of general-purpose processors (CPUs) independently developed, which greatly improved the single-core general-purpose processing performance, and realized that the CPU and motherboard upgrades did not affect the operating system compatibility. Through trial and error in the market, Loongson team realized that the main gap between Chinese CPU and foreign CPU lies in general-purpose processing performance, rather than special-purpose processing performance; It is the lack of single-core performance, not the insufficient number of cores; It lies in the lack of design ability, not the lack of advanced technology. Therefore, Loongson Zhongke Company has been committed to improving the single-core general processing performance by optimizing the design scheme, until 3A4000 completes the “make-up class” of design capability. Finally, the general processing performance of 3A4000 is equal to that of the excavator processor, the final product of AMD’s 28nm process.
CPU, the central processing unit, is the core of computer equipment, which mainly interprets computer instructions and processes data in computer software. Godson Zhongke released Godson’s new generation general CPU product 3A4000/3B4000. They are the first quad-core processors based on GS464v microarchitecture among Godson 3 series processors. Get a better understanding of the advantages of domestic processors from the perspectives of architecture and products.
structure
Compared with the previous generation GS464e microarchitecture and GS464v microarchitecture, the Loongson processor design scheme further optimizes the pipeline, improves the running frequency, and strengthens the support for virtualization, vector support, encryption and decryption, and security mechanism. Compared with the previous generation quad-core processor Godson 3A3000, the overall measured performance of the chip is about doubled. Operating system applications are binary compatible with Godson 3A3000.
Loongson 3A4000/3B4000 adopts brand-new FCBGA-1211 package, which is no longer forward compatible. Loongson 3B4000 supports multi-channel consistent interconnection with up to eight pieces of structure.
product
Loongson 3A4000/3B4000 uses the same 28nm process as the previous generation product 3A3000/3B3000, and adopts the new generation processor core GS464V newly developed by Loongson, which belongs to the industry leading new generation microstructure. The main frequency is 1.8GHz-2.0GHz, and the fixed-point and floating-point single-core scores of SPECCPU2006 both exceed 20 points, which is more than twice that of the previous generation.
By optimizing power consumption management, the working time of notebook based on Godson 3A4000 is more than doubled than that of the previous generation. The comprehensive performance of 3B4000 4-socket server formed by direct connection of CPU is more than four times that of the previous generation 3B3000 2-socket server, and the virtual machine efficiency has also increased from more than 85% of the previous generation to more than 95%.
In terms of security, Loongson 3A4000/3B4000 integrates security mechanisms on the chip, realizing the unification of self-controllability, security and reliability. It can effectively prevent vulnerabilities such as Meltdown and Spectre from the mechanism, support encryption and decryption algorithms such as MD5, AES and SHA, support dedicated secure trusted modules and state secret algorithms, and support access control mechanisms such as “shadow stack”.
Furthermore, all the functional modules in Godson 3A4000/3B4000 chip, including CPU core, on-chip interconnection bus, DDR4 memory controller and various IO interface modules, are designed independently.
In addition, all custom modules in the chip, including multi-port register file, phase-locked loop, DDR4PHY, high-speed IO interface PHY and other layouts, are independently developed. Godson 3A4000/3B4000 does not use any third-party IP except the basic design environment provided by the streaming manufacturer.

And characteristics.
The following two products of Loongson, 3A4000 and 3P4000, respectively, introduce their product features and new features.
1) Godson 3A4000
Loongson 3A4000 can be applied to Loongson desktop and notebook computers. In terms of performance, Loongson 3A4000 is clocked at 1.8GHz-2.0GHz, and each CPU chip contains four independent processor cores. The general computing performance of Loongson 3A4000 is more than twice that of Loongson 3000. The data shows that the fixed-point and floating-point single-core scores of Godson 3A4000 are all over 20, while that of Godson 3A3000 is about 10. Godson 3A4000 has a built-in 256-bit vector computing unit, and its peak computing performance for scientific computing and high-density numerical information processing is more than four times that of Godson 3A3000.
Godson 3A4000 realizes fine power consumption management, and has a built-in power consumption control core, which can dynamically adjust frequency and voltage according to the running load; At the same time, it also supports dynamic voltage adjustment, which is a leading technology in self-developed processors. According to the official, the working hours of Loongson 3A4000 notebook computer are more than doubled than that of Loongson 3A3000 notebook computer, which is optimized by operating system cooperation.
2) Godson 3B4000
Loongson 3B4000 belongs to Loongson server CPU product line and is used for multi-channel server products. Loongson 3B4000 is clocked at 1.8GHz-2.0GHz, and each CPU chip contains four independent processor cores. It supports two-way and four-way servers, that is, two or four Godson 3B4000 chips are installed on a server motherboard, and a server contains at most 16 processor cores. All CPUs are directly interconnected through high-speed bus interfaces, and share physical memory. Loongson 3B4000 specially optimizes the high-speed interconnection bus between CPUs, and the actual bandwidth of cross-chip memory access is increased by more than 400%.
Godson 3B4000 has improved its support for large memory capacity, supporting DDR4 memory, and the maximum memory capacity of 4-way server can reach 1TB. The performance of Godson 3B4000 4-way server is more than four times that of Godson 3B3000 2-way server.

Leave a Reply

Your email address will not be published. Required fields are marked *