Huawei tried to get ahead of the hardware and seize another entrance to the artificial intelligence track.
On the last day of HUAWEI CONNECT 2017, Huawei released a low-profile FusionServer G series heterogeneous computing platform, including G5500 and G2500 servers. This product is the main product of Atlas, which is built on Huawei's intelligent cloud hardware platform, and is intended to break the traditional general-purpose CPU boundary. Enable heterogeneous acceleration and improve computational efficiency. It is worth noting that Huawei mentioned NVIDIA in the technology release and said that it will jointly launch NVIDIA's GPU-accelerated cloud services based on the G-series heterogeneous computing platform and NVIDIA Tesla P100 and Tesla P4, providing users with both training and An intelligent cloud platform for reasoning acceleration.
Previously, CBN had exclusively reported that Huawei will cooperate with Nvidia in the AI ​​field and jointly build a public cloud AI platform in artificial intelligence, machine learning and deep learning. In this conference, some details of the cooperation between the two parties were disclosed. For example, the technology loaded by NVIDIA, Huawei's server can easily process 100 billion records in milliseconds.
From safe cities to smart manufacturing and autonomous driving, with the rise of AI research and applications, advances in AI and cloud computing technologies will affect every aspect of human life, and applications will be everywhere. The basis of intelligence is the ability to process massive amounts of data generated in a fully connected world. For example, a city with 100,000 cameras can generate 100 billion license plate records and 1 trillion face recognition records per year. The processing, query and analysis of scale data will pose a huge challenge to traditional computing systems.
"With the advent of the artificial intelligence era, traditional hardware can no longer meet the needs of AI and cloud technology." In a speech at HUAWEI CONNECT 2017, Qiu Long, president of Huawei IT server product line, mentioned that Huawei released a new generation of intelligence. Atlas, the cloud hardware platform, is designed to solve the data processing challenges of the AI ​​era, mainly for public clouds, AI, HPC and other scenarios.
Insiders of Huawei told reporters that hardware resources are provided on demand according to the business model. Through this platform, resource utilization can be increased by more than 50%, performance is improved by more than 10 times compared with the traditional x86 architecture, and hardware models are greatly reduced. In addition, Atlas can also generate logical servers with different resource matching relationships, second-level deployment, flexible response to business changes, and significantly shorten the business deployment cycle. Among them, NVIDIA's products play a key role.
Huawei provides information that Huawei FusionServer G5500 is a heterogeneous computing server for data center deployment. The 4U chassis supports 8 NVIDIA Tesla P100 accelerators and is compatible with the upcoming Tesla V100. With powerful GPU acceleration, it can be in milliseconds. Handle hundreds of billions of data records in a timely manner. Huawei FusionServer G2500 is a video intelligent analysis server for safe city, intelligent traffic and other application scenarios. The 4U chassis supports 16 Tesla P4 accelerators and 24 3.5 hard disks, which can satisfy the inference analysis and data storage of 256 HD video streams. .
According to the reporter, the Tesla P100 and Tesla P4 are two Pascal-based GPUs that NVIDIA launched last year. The former is mainly used to perform deep learning neural network tasks, and the latter is mainly used for image, text and speech recognition. NVIDIA has said that the Tesla P100 has a research and development cost of $2 billion and integrates 15 billion transistors on a single chip.
With the advent of the artificial intelligence era, various technology giants began the first round of layout.
The "software team" led by Internet vendors seems to have a louder volume.
For example, the innovation of artificial intelligence in the fields of image, voice, security, etc., such as Glings, Face++, Shenma Voice, etc., deep learning represented by graphic images and audio and video is a typical computationally intensive application, which gave birth to The Internet industry's need for high performance computing.
Alibaba Cloud HPC leader Wang Wei has said that Alibaba Cloud not only provides hardware and basic systems, but also HPC prepares Docker instances, scheduling systems, monitoring platforms, and training, forecasting middleware and tools for deep learning. The implication is that high-performance computing requires not only hardware facilities, but also the "soft power" of matching technology to maximize hardware performance.
Internet vendors are not at all soft on the investment in hardware. Jinshan Cloud officially launched the P3I instance of the GPU cloud server instance in early September, and called it the industry's first cloud server product using the NVIDIA high-performance computing card Tesla P4. Ali United Nvidia announced the establishment of a joint lab and opened the Nevada Kepler architecture GPU native assembler.
In contrast, Huawei's approach to “cooperating†partners and developers through hardware platforms requires long-term and sustained investment.
Dai Wei, chief analyst of Forrester, a third-party consulting firm, told First Financial reporter that from the public cloud strategy to the launch of various software and hardware innovation services, such as artificial intelligence services EI, Atlas, App Builder and DevOps Cloud 2.0, Huawei Cloud computing is based on strategic deepening and ecological expansion in multiple dimensions, while large enterprises, small and medium-sized enterprises, and developers are different groups, for service richness, stability, reliability, heterogeneous management capabilities and integration capabilities. The requirements for ecological integrity are different, and market strategies need to be formulated separately.
When talking about the difference with Internet vendors, Dai Wei told reporters that Internet companies and Huawei's DNA are different. The Internet has technological advantages and accumulated experience in its own field. Huawei has extensive experience in telecom operators and from chip to application. From the hardware to the comprehensive strength of the software, there is also the accumulation of business experience in its own field.
2222Bossgoo(China)Tecgnology.(Bossgoo(China)Tecgnology) , https://www.cn-gangdao.com