Recently, the "2021 Open Data Center Summit" was held at the Beijing International Convention Center. The theme of the conference is "Promoting High Quality and Low Carbon, Building a Better Life", and the participants exchanged and shared in-depth information on server technology trends, edge computing application scenarios, intelligent monitoring development strategies, new technology testing and evaluation standards, low-carbon green operation models, facility optimization technology research and other content around the low-carbon and high-quality development path of data centers.
Data center energy saving, need to start from the root
Why focus on data center energy savings? Because the power consumption of data centers has gone from "insignificant" to "terrible", it has only taken more than ten years!
According to the data, the number of data center racks in China increased rapidly from 1.244 million in 2016 to 3.15 million in 2019, with a three-year growth rate of 154%. The proportion of super-large data centers increased from 11.3% to 37.4%. In 2020, the total energy consumption of domestic data centers has exceeded 200 billion kWh, which is equivalent to consuming 60 million tons of coal, and it is expected that the power consumption of data centers in 2030 will be 5.1 times that of 2020. Such a growth rate is enough to describe it as horror.
Therefore, energy savings in data centers are imperative. In addition to the traditional cooling and heat dissipation optimization, the most intuitive thing is to save energy through the processor.
At the summit, Liu Hongbing, Global Vice President of AMD and General Manager of Enterprise and Commercial Business Unit in China, delivered a speech with the theme of "Technological Innovation Drives New Changes in Data Centers". Details how processors can help data centers save energy.
Hongbing Liu, Global Vice President of Amd and General Manager of Corporate & Commercial Business Unit in China
When it comes to enterprise-level processor energy saving, I believe that some friends will have doubts, after all, the current number of processor cores is increasing, power consumption, heat generation is also rising, to the data center power supply, heat dissipation have brought great challenges, and this trend is difficult to reverse.
If the power consumption of a single processor is difficult to reduce, then you can try a different angle. In 2019, from the official release of amd's second-generation epyc processor, it opened a path to build efficient efficiency. In 2021, AMD launched the third generation epyc processor, which brings the processor core size and performance to a new high, while helping to reduce TCO.
In a fixed number of cabinets, amd platform servers can achieve higher performance, not only increase the computing density, but also reduce power consumption, which is what users want to see, but also the needs of the industry. With strong performance, amd's "more than a dozen" has gradually been recognized by users, and the market share has soared!
According to the data of puget systems, a well-known foreign workstation and professionally designed machine manufacturer, the proportion of amd processor shipments has risen from 5% in June last year to 60% in June this year. Market research data for the second quarter of this year shows that AMD's share of the x86 server market was 9.5%, an increase of 4% year-on-year.
Modular chiplets power processor innovation
Liu Hongbing said in his speech: "Amd has always maintained the vitality of innovation! We are committed to making the best CPU with a dedicated spirit and providing strong computing power to the industry. ”
So, what are the aspects of these innovations? The answer: a modular hybrid chip design.
Traditional x86 processors generally use the same size of multi-core combination, the disadvantage is that the number of processor cores and the efficiency of the interconnection are difficult to improve after a certain stage. In contrast, amd epyc adopts chiplets modular hybrid chip design, supplemented by advanced 7nm process technology, can combine more processor cores, through the clever way to achieve the first 64-core 128 thread x86 processor, providing excellent io performance and efficiency.
On the booth, the third generation epyc is real
Amd epyc processor, ccx is called a unit module, generally designed as 8 cores, 16 threads, if the processor is 16 cores, then two sets of ccx are required.
In the previous generation of "zen 2" architectures, there were two sets of 16mb l3 caches in each ccx. The latest zen 3 architecture is to put the 8 cores in ccx together, sharing a 32mb l3 cache. The advantage is that with the total amount of cache unchanged, each core can use more l3 cache space in most cases.
Under the same fixed frequency, core number configuration, combined with dozens of application load test results, the iPC of the "zen 3" architecture is increased by up to 19% compared with the "zen 2"!
The data is not moved, and security comes first
"When there is a big concentration of data, it is also the time when the risk is concentrated." I believe that this has long become the biggest concern of IT people in the era of big data.
In the past two years, data thefts from large enterprises have occurred repeatedly, and hackers are constantly stealing data through more advanced tools. Amd processors encrypt memory through a separate chip, which in turn helps prevent data leakage in memory. In addition, amd has extended avx2 to 256 bits, enabling it to speed up encryption and decryption algorithms.
For virtual machines, amd's "zen3" architecture enhances the sev-es feature to limit the injection of interrupts and malicious programs, and to debug registers into the swap state. Inheriting the security features of the previous generation, amd provides cryptographic capabilities for key generation generation and management, building a hardware-level based security platform. Before the operating system kernel executes the bios code, the rom is loaded by the security processor, the bios is verified, and the key management is performed at the same time, and the boot step is started after the verification is passed. Further reduce security risks.
Simply put, from hardware to software, it can guarantee the security of virtual machines, memory and the entire system, and has been recognized by enterprise-level service providers such as vMware and AWS.
Practical application, amd big show muscle
"Many of the services we use now are based on amd epyc support. Including Microsoft Office 365, Tencent Conferencing, Zoom, Twitter and so on. At the server sub-forum, Zhou Junjie, Sales Director of Amd China Enterprise and Commercial Division, will deliver a keynote speech on "Amd EPYC Processors Make Data Centers Beyond Boundaries". In just a few short years, AMD EPYC has become ubiquitous.
Zhou Junjie, Sales Director, Corporate & Commercial Business Unit, Amd China
At present, cloud computing has penetrated into all aspects of our work and life, in order to support these complicated online businesses, cloud service providers have built many large-scale data centers around the world. So, how does AMD perform in this regard?
In his speech, Zhou Junjie introduced amd's circle of cloud service providers, including aws, Alibaba Cloud, Google Cloud, IBM, Microsoft Azure, oracle cloud, Tencent cloud, etc., covering almost all the top cloud service providers.
In terms of green energy saving, "one dozen more" has once again become a hot topic in the forum. In the AMD relational database solution, the customer replaced the original 4-processor platform with a dual-processor third-generation epyc server, and the performance was improved by 14%.
In the data analytics solution, the lower cost of the single-socket server can be described as a big success, the customer replaced the two-socket platform with a single-socket epyc server, and the performance was improved by 60%. This also caters to the theme of high-quality and low-carbon conference, so that energy saving and high efficiency can be combined.
On the server sub-forum, Zhou Junjie also revealed a message: the supercomputer frontier with computing power of more than 10 billion times will be completed this year, and it will once again challenge the peak in performance!
The frontier system design uses a new generation of HPC, AI-optimized amd EPYC processors and amd instinct compute cards. The dual-A combination of Frontier expects peak processing power to reach 15 billion flops.
It will provide simulations and modeling for Oak Ridge National Laboratory to advance the computational needs of deep interactions in weather, subatomic mechanisms, genomics, physics, and other important scientific fields.
Nowadays, the development momentum of AMD can be described as "how to soar above the sky, straight up thousands of feet". In the new era, AMD continues to apply innovative technologies to practice, providing new solutions for the digital transformation of all walks of life. In terms of energy efficiency, relying on advanced manufacturing processes and architectural design, it has greatly improved the performance of single processors, greatly leading in energy efficiency ratio, and has been recognized by cloud service providers, large enterprises and users in various industries to help them better build green data centers.