[51CTO.com original article] Recently, Huawei held the first Huawei Intelligent Computing Conference in Beijing with the theme of "Starting with the Chip, Making Intelligent Computing Pervasive". At the conference, Huawei grandly released its China Intelligent Computing business strategy, announcing that Huawei Intelligent Computing will focus on computing power, engineering, cloud-edge collaboration, and integrated solutions to build full-stack, full-scenario intelligent solutions for the industry, accelerate the industry's intelligent process, and enable the industry's intelligent reconstruction. Combined with Huawei's early chip layout of "computing, transmission, storage, and management", the reporter has reason to believe that the early accumulation of strength has been completed. Huawei Cloud, which integrates Huawei's latest Ascend series AI chips, powerful computing power, and rich intelligent computing product lines, will take the lead in entering a new era of intelligent computing and prepare to welcome the explosion of the intelligent computing market scale of hundreds of billions.
The intelligent era requires a more “matched” cloud To truly embrace the intelligent era, it is not as simple as purchasing a few intelligent computing products, nor is it to pin your hopes on the "cloud" and think that everything will be fine as long as you are on the cloud. In fact, the cloud in the intelligent era has completely different requirements for computing power, and the entire infrastructure of the entire cloud data center must be matched to adapt to the arrival of the intelligent era. In many intelligent application scenarios, it can be clearly felt that traditional clouds are stretched to meet the needs of the intelligent era: For example, CNN (convolutional neural network) is a typical scenario of deep learning. Its computing characteristics are massive matrix multiplication and addition operations. Traditional general-purpose processors cannot efficiently meet the computing requirements of CNN. In the training scenario, the parameters of the neural network need to be synchronized between multiple nodes through the network inside the data center, which places very high demands on the bandwidth and latency of the data center network. AI requires computing based on a large amount of data, so cloud storage is also required to provide massive data storage capacity and very strong IO performance. In addition, the power of AI-specific chips is far greater than that of general-purpose processors, requiring data centers to provide more powerful heat dissipation capabilities, such as using liquid cooling technology to dissipate heat for AI chips. Obviously, general infrastructure faces bottlenecks in computing, storage, networking, computer rooms, and other aspects. Cloud infrastructure must be optimized and upgraded as a whole to meet the requirements of the intelligent era. This is why Huawei has developed multiple chips and launched services such as high forwarding performance, low-latency cloud disks, and heterogeneous instances. Improve computing power at four levels and become the "first-class cloud" in the intelligent era So what explorations and practices has Huawei Cloud made to transform itself into a "top cloud" that better meets the needs of the intelligent era? Among all the active explorations, the most dazzling change is the improvement of computing power. Based on intelligent computing, Huawei Cloud will achieve a significant improvement in overall computing power in four aspects: AI computing power, innovative computing power, heterogeneous computing power, and X86 computing power: AI computing power has soared, and "inclusive AI" is no longer far away: Earlier, Huawei released the Ascend series of AI chips, which will become the core cornerstone of Huawei's intelligent computing; based on this, Huawei Cloud exports intelligent computing capabilities through various forms of AI intelligent computing instances such as virtual machines and bare metal. For example, the high-performance instance H6, which will be launched in China for the first time, will give cloud servers super-high AI computing power. Compared with the high-performance computing H3 instance, the price of H6 is 0 increase, but it provides up to 16TOPS (int8) of inference computing power, with super high cost performance, creating "affordable, easy to use, and safe to use" computing power, and accelerating the implementation of "inclusive AI". Innovative computing power empowers mobile Internet upgrades: Based on Huawei's self-developed TaiShan server for intelligent computing, Huawei Cloud provides the industry's exclusive cloud phone innovation solution, which can achieve the same instruction set as mobile Android terminals, without translation, better compatibility, and more efficient performance. With the continuous evolution of cloud phone services, Huawei Cloud can provide more possibilities for innovative applications in the mobile Internet field. Atlas creates leading heterogeneous acceleration services: Based on the Atlas hardware platform, Huawei Cloud has released GPU-accelerated cloud servers and FPGA-accelerated cloud servers, truly realizing the "hard use" of the cloud! While greatly improving performance, it reduces usage costs. Even in scenarios where storage bandwidth and latency are very sensitive, the comprehensive computing performance of the entire system can be maximized. KunLun sets a new high for X86 computing services: Huawei Cloud is equipped with KunLun servers, providing flagship bare metal services with specifications up to 768 cores, 32TB memory, and all-flash memory. This is also the industry's largest X86 computing service on the cloud. It is the only cloud server in the industry that supports both physical and logical partitions and can flexibly allocate physical computing resources on demand. Based on this, customers' resource utilization and elastic expansion efficiency are greatly improved. Based on the improvement of these four layers of computing power, Huawei Cloud provides industry-leading general computing, bare metal, heterogeneous and other service-oriented instances, converting computing power into services, providing more competitive solutions for scenarios such as AI, HPC, SAP HANA, and cloud phones, and providing the most suitable computing power services for different corporate demands, accelerating customers' digitalization process and intelligent upgrades. After being "forged" by intelligent computing, Huawei Cloud has provided more powerful cloud services to many industry customers. For example, the bare metal service based on KunLun has been successfully applied to the cloud of Harbin Pharmaceutical Group's core applications, and the heterogeneous service based on Atlas has been successfully applied to the cloud training business of a Shenzhen research institute, etc. 1+1>2 There is no doubt that with the rapid development of innovative technologies such as AI, IoT, and big data, the intelligent era has arrived. Users have put forward higher requirements for the computing power of cloud service providers, and intelligent computing will become a key criterion for cloud vendors to compete in the intelligent era. Today, Huawei Cloud and Huawei Intelligent Computing belong to the Huawei ICT family and have achieved deep integration of full-stack technologies from chips to service layers, promoting and improving each other. On the one hand, HUAWEI CLOUD realizes cloud service product innovation through intelligent computing, helping customers achieve intelligent transformation. Taking HUAWEI CLOUD's high-performance computing solution as an example, intelligent computing supports HUAWEI CLOUD and helps HUAWEI CLOUD launch the world's first high-speed network public cloud solution that supports high-speed networks. It also provides network intelligent repair capabilities for future intelligent cloud computing centers. On the other hand, the feedback from Huawei Cloud's enterprise customers has also promoted the evolution of intelligent computing, making Huawei Cloud's intelligent computing closer to customer needs. For example, when a large number of customers want to have AI reasoning services at a lower cost, Huawei Cloud promotes intelligent computing to accelerate the pace of AI chip commercialization and support a variety of deep learning frameworks and algorithms; when users have a large number of high-bandwidth demands for the network in AI training scenarios, intelligent computing is also constantly improving, and finally launched the second generation of 1822 smart network cards, increasing the supported bandwidth from 40G to 100G. The biggest beneficiaries of the deep integration of the two parties are of course customers. In the past, a model needed to be built in to monitor traffic conditions through cameras. If a user's data center wants to deploy a new model, the traditional approach is to upgrade it one by one on site, which is time-consuming and labor-intensive. Now, Huawei Cloud only needs to upgrade it on site in the data center, and then send it directly to the camera terminal through a unified architecture, and immediately obtain the latest and most advanced artificial intelligence framework application. It is not difficult to see that Huawei Cloud is opening up Huawei's 30 years of technology accumulation and products in the field of ICT infrastructure to customers in an online way. Through Huawei's full-stack innovation at the data center level in optoelectronics, chips, hardware systems, computer rooms, power supply, heat dissipation, operating systems, storage systems, networks, data processing, AI, etc., Huawei Cloud has achieved DC-level innovation and brought customers "affordable, easy to use, and reliable" intelligent infrastructure, which is increasingly playing the role of "black soil" in the intelligent world. We believe that Huawei Cloud, which has entered a new era of intelligent computing, is bound to better empower all walks of life and release more and more surging power to build an intelligent world. [51CTO original article, please indicate the original author and source as 51CTO.com when reprinting on partner sites] |
<<: WiFi stuck, maybe it's the wireless network card
In order to expand investment in strategic emergi...
[[415906]] Recently, the Information and Communic...
EMA has studied how organizations use multiple ne...
To be honest, when I first heard the words "...
In recent years, we have seen that virtualization...
JuHost has newly launched a data center in Tokyo,...
When it comes to 5G, many people's impression...
Amazon, Microsoft and Google account for more tha...
【Attention】This merchant has run away!!! Limewave...
The "old-fashioned" old boss—SDH The tr...
Ethernet is not a new technology as it has been u...
1. Dilemma of Traditional TCP/IP Network Transmis...
The previous article "Things about SD-WAN (P...
On April 12, the China Academy of Information and...
In the dynamic world of telecommunications, the a...