Recently, AMD released the world’s first 7nm data center CPU “Rome”. Subsequently, Amazon announced that it would provide related virtual machines from now on, claiming that its cost-effective ratio is 10% higher than that of Intel virtual machines. Prior to this, AMD has already scramble to release the world’s first 7 nanometer GPU before NVIDIA. Aiming at the data center market, AMD’s two attacks are tantamount to a more direct competition for Intel and NVIDIA at the same time.
Dr. Lisa Su, president and CEO of AMD, said, “Our multi-year investment in data center hardware and software roadmaps has pushed our CPU and GPU to be adopted by more and more cloud computing, enterprise computing and high performance computing customers. In the next quarter, we believe that this trend will accelerate with the introduction of CPU and GPU product portfolio, the most extensive and powerful data center with industry-leading 7nm process technology.” AMD believes that they will have a $29 billion opportunity in data centers by 2021. As the industry turns to artificial intelligence and machine learning, GPU is playing an increasingly important role in High-Performance Computing (HPC), cloud computing, super-large-scale and virtual environments. This is an important message. Apart from marking a new era of 7Nm chip technology, we can see that HPC and data center will be more closely integrated.
In recent years, HPC has been used more and more widely. In addition to the cutting-edge fields of gene sequencing, meteorological prediction and aerospace applications, Intel has given a new definition, that is, as long as there are a large number of computing and data applications, it can be classified as high-performance computing. For example, when the passengers of the drip express call, the backstage of the drip express is to use HPC to do Routing dispatch to find the most suitable driver. Another example is JD Shopping. After some female users searched for their favorite skirts, e-commerce recommended customers who had bought the goods to give them advice. These applications based on intelligent algorithms are all built on HPC. Thus, HPC is no longer exclusive to a few fields. From the earliest scientific computing to artificial intelligence, HPC has gradually simplified complex problems and applied them more in business scenarios.
Massive online business is bound to promote the development of offline data centers. Especially in the coming 5G era, the rise of super-clear video, VR/AR, artificial intelligence, machine learning and other industries provides an excellent opportunity for the construction of high-performance data centers. From server cabinet to large-scale network deployment, data center is the key to provide IT services and storage and communication for more and more network devices, users and business processes. Information from numerous network end-user devices and large data of the Internet of Things has further enhanced the value of data centers. It touches almost every aspect of human society. In history, we can hardly see any era that can produce huge amounts of data and rely heavily on these data at the same time. . As a relatively leading industry technology, HPC will have a more profound impact on data centers, so it is urgent to build a new generation of data centers and improve the performance of physical servers in an all-round way.
In previous articles, I have introduced HPC and cloud computing in detail. It is mentioned that HPC is more powerful than cloud computing because of its outstanding performance of physical architecture, such as clustering and parallel computing of CPU and GPU, which greatly improves the ability of HPC data center in data acquisition and data mining. In addition, besides the requirement of large bandwidth, the requirement of low delay and low power consumption is paid more attention by data center builders. At the same time, HPC data center will use more cloud servers to support the environment of physical machine and virtual machine, and realize the unified management and allocation of hardware and software resources. According to Cisco’s forecast, from 2016 to 2021, the number of global super data centers will grow from 338 to 628, accounting for 53% of all installed data center servers by 2021. The improvement of infrastructure and the rapid growth of user traffic will strongly promote the expansion of data centers.
In such a huge market, Active Optical Cable (AOC) is the best solution for data centers, server clusters, network switches and other high-speed data transmission, high-performance embedded application communication needs (such as HPC). Traditional copper Direct Attached Cable (DAC) can not satisfy higher performance applications, because it has a lot of drawbacks although its price is lower. Brad Smith, senior vice president and data center interconnection industry analyst at LightCounting, a market research company, said that, “The main application markets will continue to use HPC and supercomputing markets based on InfiniBand protocol. AOC’s applications in data centers over the past few years have been used to connect switch layer using Ethernet protocols. For telecommunication, long-distance DWDM and router interconnection applications at the local end have also begun to grow.”
For the short range intraconnection application of HPC data centers, Gigalight’s typical products include 10G SFP+ AOC, 25G SFP28 AOC, 40G QSFP + AOC and 100G QSFP28 AOC series. The AOC products have high reliability because they are not exposed. In addition, the high frequency EMI signal of AOC is processed in the pluggable module, so the anti-interference degree is better than DAC. Lighter weight and smaller cable diameter are also the advantages of AOC over DAC, which also meet the requirements of high density optical interconnection.
Compared with traditional data centers, HPC data centers need to support higher density device connection. However, high density brings high power consumption. The maintenance cost and risk cost caused by power consumption are always a headache in the industry. For the next-generation HPC data centers, Gigalight has launched a series of 200G/400G PAM4 AOC products, using DSP and PAM4 technologies to achieve higher transmission parameters. In the end of 2018, Gigalight(www.gigalight.com) will also introduce 2nd generation 200G/400G AOC products with higher cost performance.
