Jingdong cloud with the next generation of AI server can support future production of nearly five years, technology evolution

5G, big data, edge computing development is driving innovation in the field of cloud data centers, server design, form, function also simultaneously upgrade. Recently, Jingdong cloud and AI research since the completion of the next generation of server prototype production, production is expected to soon be listed. Jingdong cloud and next-generation server AI has significant advantages in terms of computing performance, memory bandwidth, the system architecture. The hardware system is modular in design, and optimized for power, cooling, management, energy and other infrastructure, "high stability, high flexibility, high efficiency, low cost" is designed to enable server-wide flexible deployment scenarios, to meet the enterprise wide demand for different application scenarios cloud server, especially in high-performance computing and artificial intelligence will play a greater value.

In the cloud computing era, due to the increased demand and enhance its cluster model, the traditional whole cabinet server energy consumption, long deployment cycles, operation and maintenance and other defects become more and more complex, which the server's stability, flexibility, a higher efficiency requirements, but also to promote the development of centralized data centers, large-scale direction, more and more large and very large data centers, server machine cabinet came into being.

In order to make the whole cabinet for the broader market, Jingdong cloud AI combined with the user's needs and pain points, through the "standardized, modular, flexible" and can be combined to achieve the whole design concept of flexible deployment scenarios, Jingdong R & D under cloud with AI generation server (hereinafter referred to as "the next generation of server"), which along with the whole cabinet server cost, rapid deployment, low power consumption, ease of management and other aspects of the advantages of centralized power and innovation combined with traditional stand-alone server power supply two power supply way, combines the dual advantage of air-cooled and liquid cooling, fully modular design, allowing users like building blocks with a product configuration more fit the actual application scenarios demand, easy "on-demand" to help users achieve full flexible deployment scenarios.

Jingdong cloud AI and hardware R & D base, research director Chen Guofeng opinion, the emergence of the next generation of servers, infrastructure development is in line with the market, breaking the limitations of the application, the ecology of the whole cabinet to a much broader, more compatible more flexible development, server architecture makes more applicability, regional and wider applications, more reliable equipment operation, lower operation and maintenance costs, providing a full range of options for users of cloud services IT facilities.

As a 19-inch standard machine cabinet servers, next-generation servers to achieve a high elasticity and regularization, including higher power density, improved power and cooling, better compatibility, more resilient and flexible deployment, solve large-scale and large scale CSP needs of the data center to better support users secure access to cloudy environment.
Jingdong cloud with the next generation of AI server can support future production of nearly five years, technology evolution

[Standard machine cabinet and cabinet switch back and
improve flexibility]
mainstream machine compared to standard rack cabinet, in terms of structure or system architecture are very different, such as the size of the cabinet, fan and centralized power management, resulting in limitations of the whole cabinet to deploy, difficult to achieve flexible migration. Jingdong cloud generation with the AI server 42U, 19-inch standard rack design units, nodes independent cooling, versatility, the whole realized arbitrarily switch cabinet and standard rack, the user can use U disk "Plug ", dramatically reducing the barriers to entry and investment in data center costs.

[ "Hard front, rear fan"
a new system architecture to protect the hard disk]
on the frame of the cabinet, next-generation server design front IO, the electric and electronic separate operation and maintenance of the data center and collocation cabinets pre-processing trough , rear of the cabinet without any cables, ensure that all operation and maintenance work can be carried out in the cold aisle, the operation more efficient. IO server encounters the current standard cabinet, the user need only remove a fan replaced after IO modules may be implemented on the line, while the server at the rear end of the entire fan, the fan and hard disk greatly widened the distance, it reduces the noise impact and vibration on the hard drive.
Jingdong cloud with the next generation of AI server can support future production of nearly five years, technology evolution

The modular architecture of the power supply [] matches the diverse needs of
next-generation server in the power supply design is also modular, independent power supply node, using the standard of CRPS, and in conjunction with conventional centralized power supply two power supply stand-alone server, the user can choose unified architecture of centralized power supply, centralized power can also be replaced one pair PDU, it is transformed into a traditional stand-alone server power supply, in order to meet the needs of different users.

【风液混合散热实现高密度部署】
随着CPU功耗越来越高,AI场景更加普及,服务器整个系统热岛现象明显,受限于结构空间,只能加大系统风量,提高风扇转速,伴随的就是噪声高,功耗大,整体PUE过高,面对这些问题,风冷已经不足以满足目前的制冷需求,甚至散热已经制约了服务器和数据中心的发展。京东云与AI综合考虑机房运维、系统运维以及上线风险等方面,设计风液混合散热方案,散热能耗降低60%,实现更高功率密度的部署。
Jingdong cloud with the next generation of AI server can support future production of nearly five years, technology evolution

s此外,下一代服务器采用的大部分模块化设计,不管是风扇、电源、加速卡,还是前置的IO模块,都是用市面上已有的标准部件,可以在资源上实现快速交付。在架构设计上,下一代服务器不仅能兼容传统CPU架构,还可以支持针对AI计算的GPU和各种异构ASIC架构,前瞻性的高兼容架构,让整个服务器的性能得到大幅度地提高,可以支持未来3-5年的服务器产品演进,同时赋能数据分析、人工智能等各类应用,为人工智能与高性能运算做好最充足的准备。

【云+AI完美组合打造专有云解决方案】
凭借在云计算方面的良好表现,下一代服务器以高效、低耗、灵活等特性,最大效能发挥了GPU的强大处理性能,满足AI深度学习,科学计算,视频处理,图形可视化等多种应用场景,并完美支持Caffe、Torch等业界广泛使用的深度学习计算框架,同时也为超级计算和云计算应用提供一体化交付服务,在全面提升运维效率的同时降低了运营成本。

对于京东云与AI来说,下一代服务器交付的不再只是软件和基础设施,而是一整套服务器交付的标准和落地,预计很快量产上市。在云服务方面,京东云与AI也积极贡献自己的创新力量和业务能力,现推行的JD Cloud Stack专有云产品,可以提供业务上云、云原生架构设计、数据迁移、机房规划、网络设计、部署交付、安全服务、应急响应等多种服务和支持,为用户打造完整的专有云解决方案。

"The development of enterprises in today's era faced with opportunities and challenges of the digital wave brought about, companies need to be able to meet the innovative technology of different application scenarios for office. In order to more effectively help the way businesses stabilize the digital transformation, Jingdong cloud and AI constantly infrastructure solutions and services for innovation and expansion. since the official business, and Jingdong cloud AI short time that Chinese cloud computing market like gangbusters, China made public cloud IaaS market ranked ninth. today, we launched a new generation of servers also it is confident that it will continue to be the industry's leading technology and innovative optimized solutions to promote the development of the industry, help Chinese customers to open a new digital future. "Chen Guofeng representation.

Guess you like

Origin blog.51cto.com/14658489/2467813