Shulou(Shulou.com)06/02 Report--

In the past two years, artificial intelligence is undoubtedly the hottest topic, and more and more enterprises begin to devote themselves to the research and development of artificial intelligence software and hardware.

Behind artificial intelligence, a neural network is established by deep learning and simulates the human brain for analysis and learning. it imitates the mechanism of the human brain to interpret data and make machines think and judge. However, the parallel computing needed for deep learning is not the strength of CPU, so the high-performance CPU+ GPU coprocessing accelerated server is becoming the indispensable basic hardware to realize deep learning.

With the advent of the era of artificial intelligence, traditional hardware can no longer meet the needs of AI and cloud technology. Therefore, to find a more efficient professional AI computing platform has become an urgent need for many enterprises in the field of AI. The latest H3C UniServer R5200 G3 server launched by Xinhua three is just such a platform.

Excellent computing performance

H3C UniServer R5200 G3 server is a new generation of GPU server customized by Xinhua San for artificial intelligence and high performance computing. It adopts standard 4U rack design and supports two latest Intel Xeon Skylake processors.

In terms of CPU/GPU architecture design, the biggest feature of R5200 G3 is that it provides users with the scalability of up to 10 double-wide GPU or 20 single-width GPU. In order to give full play to the efficiency of several thousand computing cores of GPU and avoid the bottleneck of data IO, R5200G3 optimizes the CPU/GPU communication link and adopts 1:8 double-link communication to ensure efficient data exchange between CPU/ GPUs and GPU, so that R5200G3 has the industry-leading stand-alone computing power.

H3C UniServer R5200 G3 provides 21 full-length PCI-E 3.0 extension interfaces, which can install 10 double-wide GPU cards at the same time in x16 mode or 20 single-width cards at the same time in x8 mode (another PCI-E 3.0 interface for expanding high-speed cards). Different configurations correspond to the two main application scenarios of training and reasoning in machine learning.

When paired with the current state-of-the-art NVIDIA Tesla V100 computing card, the whole machine can provide up to 1120 TensorTFLOPS performance in 4U space. If an ordinary 2U server is used, users need 5 servers and 10U space to achieve the same theoretical performance.

High-speed network communication

R5200 G3 can support both 100Gb high-speed network card and Direct RDMA communication, and has excellent performance in deep learning and training scenarios of large-scale multi-computer parallel computing. In 8GPU\ 16GPU\ 32GPU parallel training, R5200G3 and 100Gb networks are used to achieve nearly linear performance growth.

High scalability

Although H3C UniServer R5200 G3 itself can provide strong computing power, enterprises usually use more GPU servers to cluster to achieve higher performance in larger-scale machine learning, high-performance computing and other application scenarios. To this end, Xinhua 3 provides a powerful network option for the H3C UniServer R5200 G3, and users can choose between 10Gb Ethernet and 56/100Gb Infiniband networks to achieve higher interconnection capabilities.

In terms of disk storage, the H3C UniServer R5200 G3 provides 14 SAS or SATA SFF hard disk bits that can be installed with HDD or SSD, providing sufficient storage space while meeting extreme computing requirements.

Reliable, efficient and intelligent

It can be said that H3C UniServer R5200 G3 has reached the peak of computing performance per unit volume in current commercial products, allowing users to achieve higher performance with minimal space and hardware overhead, thus avoiding the costs of expanding data centers and building large networks.

Of course, in order to ensure the continuous and stable operation of the server in the fully equipped state, H3C UniServer R5200 G3 also adopts a redundant design in the cooling and power supply part.

In the cooling section, the H3C UniServer R5200 G3 uses a 5x1 redundant fan design with hot-swappable support, which not only provides sufficient air volume for CPU and GPU to meet the cooling requirements, but also leaves sufficient heat dissipation margin for accidents that may occur.

In the power supply part, the server is also equipped with four 1600W hot-swappable 80Plus platinum power supplies, and the redundant configuration of Nissn ensures that the server can obtain extremely high energy efficiency and abundant power supply under any circumstances.

In the aspect of intelligent management, in the heterogeneous computing environment with GPU as the main computing unit, the traditional management tools are difficult to deal with perfectly, so Xinhua 3 also equipped the server with the management software-AIOS, which is specifically for the GPU heterogeneous computing platform.

AIOS can not only achieve GPU load, work status monitoring and other basic functions, but also achieve GPU resource scheduling based on the mainstream cloud and container environment, so that enterprises can use huge GPU resources more flexibly in multi-tenant environment, so as to achieve more comprehensive heterogeneous computing management.

Basic parameters:

It is worth mentioning that this new generation of GPU server products customized for artificial intelligence and high-performance computing has been fully recognized by the financial industry.

In the construction of the national artificial intelligence platform of a bank, it is planned to purchase about 200 GPU servers. As a large-scale artificial intelligence platform project in the financial industry, it has attracted the active participation of almost all domestic server manufacturers. Xinhua III artificial intelligence solution has been successfully shortlisted after fierce competition, and the bank will purchase more than 50 R5200G3/R4900G3 servers.

In view of the great potential of artificial intelligence in automated business analysis and automated process processing, an insurance company in China continues to try to use artificial intelligence technology to improve the management efficiency of the insurance industry. Xinhua 3 artificial intelligence products and technologies are selected in the construction of artificial intelligence platform, including 8 R5200G3 servers.

Summary

Artificial intelligence is not only one of the most frontier applications of computing science, but also a major challenge to the traditional hardware computing platform. With years of accumulation and huge investment in server hardware design, Xinhua 3 launched the H3C UniServer R5200 G3 server for the most demanding machine learning scenarios in artificial intelligence applications.

This product not only represents the top level of today's server design, but also represents the concept of "server design scene". Through full-dimensional innovation in internal space, power supply, heat dissipation and management, Xinhua 3 H3C UniServer R5200 G3 perfectly solves all kinds of problems in the application of machine learning, making it an ideal hardware platform in the field of machine learning.

To quickly understand the latest developments of Xinhua three products and technology, please follow the official account of "Xinhua three Commerce"!

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.