Since returning to the data center processor market in 2017, AMD has never concealed its ambition to make a difference in this field. Whether it is the generation of 霄龙EPYC's Naples, or the second generation of Rome, and the future of Milan and Genoa, it represents a powerful metaphor of “empire” and “rejuvenation”.
AMD announced at the AMD Greater China Partner Summit in Beijing a week ago that it announced that the second-generation Snapdragon EPYC server processor Rome has officially entered the Chinese market. The vast market space, strong market demand, the industry's desire and expectation for 7nm will undoubtedly bring more opportunities to AMD. But after all, "Rome is not built in a day", AMD intends to return to its former glory, continuous product iteration, regain customer confidence and market reputation, as well as a complete ecosystem, etc., all need to be carefully crafted in time, while AMD still needs to face A counterattack against Intel, possibly from a competitor.
But at least from the current point of view, in the data center business, AMD has a clear idea, fierce play, and implementation is in place. Under the "window period", relying on the various auras of advanced process blessing, AMD is beginning to build a 7nm "Roman Empire". .
The king of cost-effectiveness continues to gain confidence
AMD is quite "distracted" on Rome. It was the first publicity at the partner summit held in Beijing last year. It was officially unveiled at the Taipei Computer Show in May this year, and then released in San Francisco in August this year. The X86 server chip of the 7nm process is highly anticipated by AMD. At the time of the 50th anniversary of AMD in 2019, Rome's appearance was quite a gift.
From the performance point of view, Rome has a number of indicative data, very able to play. The world's first server-class CPU supporting PCI-E 4.0 technology, with up to 64 cores in the industry. The new Zen2 architecture delivers a 15% overall performance improvement over the previous generation, with twice the performance and 4x floating point performance of the previous generation of Naples, as well as innovative small chip system-level packaging and security features. As the most powerful chip in the current X86 server chip, Rome has more than 100 record performance.
At the same time, the 7nm process also further reduces power consumption and cost, making Rome a very strong price/performance advantage compared to Intel's products. Whether it's AMD's demonstration of a single 64-core Rome vs. Intel dual 28-core 56-thread Xeon Platium 8180M, or a large number of third-party test results, Rome can win.
In addition, Rome can reduce total cost of ownership (TCO) by up to 50% under a variety of workloads, and Rome's price can reach half of Intel's products in terms of price.
In IC design, architecture and process are key indicators that determine product performance. With these advantages, AMD scored a quarter of the server CPU market share in 2003-2006, and then Intel offered "Tick-Tock" "The strategy has regained the lost market.
In the past ten years, AMD has never had such a competitive product in the server CPU field. Now, the initiative has returned to AMD. On the AMD, AMD has achieved Intel's leading position in the process technology for the first time in ten years. I have always chosen to gamble on 7nm and succeed, and Intel's delay in the process has also given AMD an excellent opportunity for development.
The rapid development of the semiconductor industry has led to the company's strategy is extremely important, and can not be lost. After AMD announced its return to data center business and X86 architecture in 2017, it has formulated its first five-year plan. Through the continuous evolution of the product line, it hopes to complete the triple jump of “enter, catch up, and surpass”.
Today, the 14nm "Zen" architecture of the first generation of Naples, 7nm "Zen2" architecture of the second generation of Rome has been successfully launched, according to AMD, the current 7nm + "Zen3" architecture of the three generations of Milan has been completed as scheduled, four generations of Genoa is designing as planned in.
A clear roadmap evolution, stable product iterations, and strong execution are key to ensuring AMD's success. According to the collection micro-network, in the past three years, AMD is advancing these plans as scheduled, and the time error will not exceed one quarter.
What is commendable is that, unlike Intel's 10-year-old server CPU performance, the “squeezing toothpaste” has evolved slightly. The 霄龙EPYC series has achieved a significant performance improvement for each generation of products, which is very suitable for the current high computing power. The required market and industry status.
This is also the second important principle of AMD's CEO Su Zifeng's focus on the ultimate performance product, namely efficient and powerful execution. Through continuous evolution, the corresponding products are launched at the process process nodes as scheduled, and the commitments are repeated. On the basis of ensuring performance, the product roadmap is guaranteed to be executed stably and on time.
In this way, AMD hopes to bring disruptive changes to the data center, and more importantly, to provide customers with a long-term stable route, giving the market trust and confidence in AMD products, and continuously obtaining customers.
The evolution of the two generations of products has made AMD see the hope of recovery in the market. According to research firm Mercury Research, AMD's server chip market share increased to 3.2% in the fourth quarter of 2018. According to DigiTimes, AMD currently has a market share of approximately 5%. After the release of Rome, the general market expectation shows that AMD's market share on server chips will reach 10% by the end of 2019. Compared with less than 1% when returning to the data center business two years ago, it has already grabbed 10 times market share.
Reinventing the data center to reject "turtle growth"
The rise and boom of emerging industries such as cloud computing (especially in the field of edge computing), big data, artificial intelligence, and 5G have put new demands on the speed and amount of data processing, bringing new opportunities to the data center industry. Modern data centers are now facing more and more challenges, such as the pursuit of computing power, lowering total cost of ownership, increasing size, security and privacy.
So modern data center services require leading architecture, performance, scalability, and security with more powerful processors. However, in the view of Zhou Junjie, sales director of AMD China Commercial Division, in the past few years, the performance of mainstream two-way servers has increased by only 15% per generation, and the price has increased by 15%. This is almost the same as “turtle growth”. He believes that AMD is re-setting the new server CPU standard, and performance improvements are measured by doubling.
In short, AMD hopes to redefine the standards of modern data centers or to promote the transformation of modern data centers. AMD believes that a sustainable CPU, new GPU architecture, a new chip and system integration solution, and an open ecosystem will drive this transformation.
Therefore, this idea can be seen both in the design of Naples and in the parameter specification of Rome. Such as the continuous evolution of the "Zen" architecture, the innovative "chiplet" chiplet design, 2x performance improvement and up to 50% TCO reduction.
This is in line with the temperament of AMD's subversives, and of course it is in its own market position, the choices that must be made by Xiaobo, and of course based on its technological innovation and deep accumulation.
In the data center business, Snapdragon EPYC is mainly targeted at three major application scenarios: cloud services, enterprise-level, and high-performance computing.
The primary motivation for cloud service providers and enterprise customers is to reduce procurement costs, while also seeking alternatives from Intel. For the high-performance computing field, there is an infinite pursuit of computing power, and AMD can well meet this demand. At present, AMD has cooperated with mainstream manufacturers in almost all of these fields.
In the field of cloud services, you can see that Microsoft, Google, Amazon, Tencent, Baidu, Twitter, etc. have already launched related products based on Rome or are working together, and many of these vendors are also Intel customers.
Cloud service providers are the areas that AMD relies heavily on. The overall server market is the cloud and data center. With the deployment of 5G and the Internet of Things, more demands on the data center will drive the main growth of cloud servers.
Tencent Cloud has already reached an in-depth cooperation with AMD. It is a very good start. As the first cloud manufacturer in the country to reach one million servers, and the core business in the next two years, cloud manufacturers will bring AMD growth. chance. At present, Tencent Cloud has launched a new generation of cloud server instance SA2 based on the Rome processor. The overall performance has increased by more than 35% compared with the previous generation, and the price has been reduced by 30%.
At the enterprise level, it is more sensitive to cost and space. With online retail business as an application case, based on Rome's flagship product 7742, the price/performance ratio has increased by 4 times. The original enterprise Java application deployment requires 60 2-way servers. Requires 33 servers, reducing costs by 44%, reducing the number by 45%, and improving performance by 83%.
In terms of high-performance computing, including French IT giant Atos, Finnish IT Science Center (CSC), and French national high-performance computing organization GENCI are all working on Rome-based cooperation, and also set a number of performance records in the field of high-performance computing. AMD is also working with Oak Ridge Labs in the United States. It plans to deploy the world's most powerful supercomputer in 2021. The target computing power will exceed 1.5 exaflop, which will be the current global super-calculation first to 100th computing power. Sum.
China's market "the right medicine" to enhance recognition
Today, from communications, the Internet, artificial intelligence to smart manufacturing, big data, and shared economy, the digital economy has become the world's most important industrial base, business model and new economic form.
China's digital economy is also in a period of rapid growth. China still maintains the status of the world's second-largest digital economy (the first in the US). According to figures released by the Shanghai Academy of Social Sciences, China's digital economy has grown at the same rate for three consecutive years. First, it is 21.5% in 2016, 20.3% in 2017, and 17.6% in 2018. In 2019, it is expected to increase by 15% (6% in the US).
The Chinese market has become one of AMD's data center business strategy priorities, with strong market demand and broad market space. For example, in the field of high-performance computing, the organization expects global scale to reach $74 billion by 2021, of which 25% is in the Chinese market.
According to Scott Aylor, AMD's global vice president and general manager of server business, the thirst for computing power and cost is a common challenge in the global data center field. In terms of performance, innovation speed and cost performance, Chinese customers will reflect More prominent.
“China is the most active market for innovation. China has been at the forefront of the world in terms of mobile payments and 5G. New applications are constantly emerging, and new competition will be triggered. AMD is a provider of basic CPUs. At this time, the launch of the product met a variety of needs." Scott Aylor told the micro-grid reporter.
Liu Hongbing, vice president of AMD's global vice president and general manager of the Greater China Business Unit, said in an interview that today's Chinese customers are becoming more "smart", from the past only concerned about how many machines to buy, to how much power is required per watt. There is more focus on cost and energy savings because they need to grow faster while reducing energy consumption. So from 14nm to 7nm technology, AMD provides more computing power with less power consumption, all of which is to meet the needs of Chinese social technology development very quickly and better.
So how do you sell your products better? According to John Taylor, AMD's global vice president and chief marketing officer, the first is to promote the industry's attention through the promotion and promotion of EPYC's performance and other advantages. The second is to train professional sales staff to be more directly involved. The purchaser of the decision-making party is docked. The third is to strengthen digital marketing capabilities to better reach the decision-makers in the enterprise to make IT professional decision-making and cloud migration, so that they can understand that AMD can now provide more choices through EPYC products. Using this as an entry point, combined with successful cases, we will introduce these decision-makers to the more superiority of EPYC and let them gradually accept it.
At this year's Greater China Partner Summit, Tencent Cloud, Dell Technology Group, Xinhua III, Lenovo, VMware and other industry chain partners announced the latest products and solutions based on the second-generation AMD EPYC processor, which has officially landed in Greater China. District market. Scott Aylor said that the market acceptance of the second generation of Opteron servers is very good, whether it is in the cloud or OEM channel partners.
At the end of August, Lenovo officially released two server products equipped with EPYC 7002 series processors, SR655 and SR635. According to Lenovo's test results, the new products can help customers reduce hardware operation costs by 46%, while software costs can be reduced by 72%. .
"The result of the comparison is based on a standard 320 virtual machine platform. The data provided by Lenovo shows that the number of traditional two-way servers and new server is 5 to 9, and the number of CPUs is 5. The pair has 18 pairs. The license has been reduced from 18 to 10, which originally cost $400,000, and now only needs $180,000," said Huang Shan, director of product management at Lenovo Group DCG China.
A server manufacturer told Ji Wei. The customer is very interested in the second-generation Opteron server, and its multi-core architecture is very suitable for mainstream applications such as virtualization and high-performance computing (HPC), and the AMD platform is also obvious in terms of price. The advantages. Many Chinese customers are testing the AMD platform and the results are very positive.
“Especially for Internet customers, the biggest challenge is to build a data center that cannot keep up with the speed of the server. When the computing power is high, there is no place to buy it, so it needs to be supported. After the performance of the single machine is improved, the customer enjoys more. More resources, no need to buy more land, electricity, maintenance costs, etc. are also correspondingly reduced, which is crucial for users to cope with the rapid development of applications in the future," the source said.
Building a real challenge of 7nm industrial ecology in the next year
After solving the performance, cost, and continuous evolution, AMD made good progress after returning to the server market. But still need to face two key issues.
The first is ecology, which is also the most concerned issue of the outside world.
"AMD has lost a voice in the data center market for a decade, and Intel has built a complete ecosystem. Today, 95% of the X86 ecosystem is Intel, at this point AMD