The design concept of excellence first originated from architecture. It requires designers to fully consider the structural functions under extreme design in terms of building structural loads, and emphasizes that building structures must meet safety, reliability, and durability. Nowadays, this design concept has been integrated into all walks of life, among which the technology industry is a typical representative, and the design of server products is no exception.
For servers, we can look at it from three aspects: first, when designing products, design different forms of products based on different application scenarios to meet business needs; second, when selecting components, we must consider how to maximize performance and quality. guarantee; finally, the product structure and components are optimized to enhance the ease of use and reliability of the product while meeting the expected performance.
The new G7 computing power platform released by Inspur Information this year adheres to the design concept of excellence and uses the most advanced computing, storage and interconnection technologies to achieve full scenario coverage, providing full-scale intelligence from devices, complete machines to data centers. management.
Better performance comes from putting the system at the core
The architect Kaiji Ji once said: "Architectural design cannot just focus on one's own design, but must be in harmony with the style of the entire city." Design with the whole as the core is the beginning of building a classic foundation... As the computing power platform carrying data processing in the data center, the server needs to be designed from a global perspective to maximize performance, including solving architectural design, A series of issues such as core components, high-speed interconnection, and thermal design.
Inspur Information G7 computing power platform insists on taking the system as the core and has carried out a lot of innovation work at the circuit level, component level, firmware level and system level. Taking the research and development of a general server as an example, the entire system needs to go through more than 30 development processes, use more than 150 processing and manufacturing processes, strictly control the quality of more than 280 key process control points, and also need to achieve integration with software and Application optimization and adaptation issues.
The starting point of system design is to ensure power supply. The G7 computing power platform has extreme power supply capabilities. It uses third-generation semiconductor silicon carbide and gallium nitride switching elements, and the power power density is increased to 100W/in³, accounting for the highest 20% to 50% When the power supply is under load, the automatic identification and adjustment conversion rate can reach 94%, increasing the conversion efficiency by 4% to 6%.
With a powerful power source, the key factor affecting product performance is the heat dissipation capacity of the entire machine. The G7 computing power platform provides a variety of heat dissipation solutions. The traditional air cooling mode includes fan structure, heat flow air guide technology, intelligent fan control technology, EVAC heat dissipation technology, The overall machine architecture has been optimized, and complex heat dissipation control has been used to solve the problems of wind sources, wind flow fields and high-power components. The heat dissipation efficiency of the whole machine has been increased by more than 24%, and a single device can save nearly a thousand kilowatt hours of electricity a year. Faced with more stringent demands for PUE, the G7 computing platform supports innovative technologies such as cold plate liquid cooling and immersion liquid cooling. Cold plate liquid cooling covers high-power-consuming components such as CPUs and memories, and liquid cooling covers the power consumption of components. Accounting for more than 70%, it can help the PUE of the data center drop to less than 1.1.
In terms of memory optimization, the storage module E3.S of the G7 computing platform can achieve remote memory expansion through CXL. The latency performance is basically equivalent to the cross-processor memory access latency, which meets the problem of insufficient memory expansion in cloud computing virtualization scenarios. In terms of network interconnection, the G7 computing power platform supports Multi-Host network cards, which not only allows multiple CPUs to connect to a single network card at the same time, achieves multi-CPU network IO Balance, reduces cross-CPU access delays, and supports interconnection between multiple systems. Each computing and storage node can be interconnected with the outside world through a single network card, and the data center performance can be improved by more than 150%.
It is precisely because of the above design that the G7 server won the SPEC CPU, SPEC jbb dual-socket, and four-socket performance championships in the latest evaluation, achieving a comprehensive upgrade in the performance of different processor platforms and different types of products, becoming a multi-computing power platform performance benchmark.
Scenario-based design creates the strongest product portfolio
The essence of architecture is to serve functions. Opera houses and office buildings have different application scenarios and present completely different architectural forms. The same is true for the construction of data centers. The construction of data centers for different users faces different challenges, such as insufficient computer room space, Explosive growth in data volume, insufficient operation and maintenance personnel resources, and tight expansion demand cycles. Facing the needs of different business scenarios, the G7 computing power platform is optimized from multiple aspects such as computing density, storage density, and IO expansion.
For example, the development of big data technology has allowed data center users to face differentiated storage scenarios such as distributed cache and hot storage, which have different requirements for server computing performance, storage density and IO.
To this end, Inspur Information has launched a storage-optimized product - NF5266G7, which can cope with the pressure of increasing computing performance, storage density, IO expansion and other aspects. It can support 2 latest processors in 2U space, while providing a maximum of 600TB and Hundreds of storage configuration solutions allow big data applications to reduce CPU investment by 50%; in the face of computing-intensive scenarios, the G7 computing power platform can provide 2U space 4-node products, and the computing power of a single cabinet can be increased by 4.2 times; For I/O expansion needs, the entire G7 computing platform supports dual OCP and Multi-host network card expansion, supports online hot maintenance, etc., and improves performance by more than 10%.
The G7 computing power platform supports 8 types of mainstream general-purpose processor chips and 51 types of accelerated computing chips, creating products including general computing, AI computing, high-end computing, application optimization, complete cabinets, all-in-one machines, etc., and is equipped with Inspur KeyarchOS operating system. Realized server software and hardware collaboration system optimization.
Strive for excellence and stably output surging computing power
The most basic requirement of architectural design is safety. Structural safety is the ability to prevent damage and collapse. It is the most important quality indicator of structural engineering and is reflected in all aspects of construction such as materials, structure, fire protection, and electrical. Inspur Information has always been the same when it comes to product creation, insisting on integrating the design concept of excellence into each product to create high RAS characteristics.
The G7 computing power platform adopts the "three-layer protection model for the whole machine" to improve product reliability layer by layer from physical hardware redundancy, system derating design and software system protection design. Redundant design of key components, such as multi-channel power supply for CPU, N+1 redundant design of fans, PSU and other components, dual ROM design of BIOS and BMC, etc. After one component fails, the redundant component This ensures that system operations will not be interrupted.
The derating design of the system means that when the power supply capacity of the cabinet is limited or the power supply of a single PSU is insufficient, Inspur Information's unique dynamic Power Capping technology provides overload real-time response measures to ensure the safety of the system power supply, ensure business continuity, and achieve the motherboard MTBF trouble-free working time. More than 24W hours. Behind this is the rigorous selection of every capacitor on the motherboard. Through tens of millions of tests and simulation experiments, the component layout and heat dissipation strategy are optimized to reduce component stress and slow down aging, thereby extending product life and avoiding different business scenarios. Abnormal system operation problems caused by differences in computer rooms; software system protection is another evolution of reliable computing power. In order to support hot-swappable protection of more components, the G7 computing power platform supports PSU, fans, hard disks, OCP network cards, etc. The Efuse protection mechanism of each component can achieve millisecond-level direct fusing when an abnormality occurs, providing strong protection for the system.
In addition to rigorous product-level design, Inspur Information's own ace laboratory - "Pengpai Laboratory" conducts more than 2,000 tests on each product, including more than 30 extreme tests, developed process documents, production quality control Standards are distributed to factories around the world through information systems and uniformly executed through intelligent manufacturing systems, thereby achieving global high-quality delivery.