AMD's second-generation EPYC processor not only uses the advanced 7nm production process, but also based on the new Zen 2 architecture, and its core specifications up to 64 cores and 128 threads, supports 128 PCIe 4.0 channels, and provides very reliable security from the hardware level. Sex. From our test results, the actual performance of AMD's second-generation EPYC processor has been greatly improved based on the previous generation. So how does AMD view the market competition after the second release of the EPYC processor is officially released, and how does the second-generation EPYC processor enter the Chinese cloud computing market? Mr. Scott Aylor, AMD's global vice president and general manager of the data center product division, and Mr. Liu Hongbing, AMD's global vice president of commercial sales in Greater China, shared their views.
▲ Mr. Scott Aylor, Vice President of AMD and General Manager of Data Center Products
MC: Previous data showed that AMD's market share in server chips has increased from 0.8% in 2017 to 3.2% at the end of 2018. After the second generation of EPYC products are released, do you have any expectations or a quantitative target for the market?
Scott Aylor: Dr. Lisu Su said in the first quarter of this year that our goal is to achieve double-digit market share in the coming quarters. We are very much looking forward to the second generation of EPYC processors to be a boost to our market share growth.
MC: What are the advantages of the second generation AMD EPYC processor in the Chinese cloud computing market?
Scott Aylor: When we look at the second-generation AMD EPYC processor, we can see that its chances in the cloud are very large, because the core of the second-generation AMD EPYC processor is very large. This means that if users choose to use it to build a server platform, it can run more virtual machines than the previous generation, and the energy efficiency ratio is better, and the TCO is more affordable. We are also very much looking forward to the cooperation of these large-scale cloud service providers in China and some potential second-tier cloud operators to the second-generation AMD EPYC processor.
Liu Hongbing: I add that the Chinese Internet market is relatively mature today. A large number of Chinese Internet customers are paying more and more attention to TCO and revenue. At the same time, the public is also paying attention to the income of these Internet companies. This is the second generation of AMD EPYC. The advantage of the processor.
MC: The Chiplet used in the second-generation AMD EPYC processor is a very successful design. What is the key to its success?
Scott Aylor: In fact, because the Chiplet design is mainly focused on the technology of Inter-Connect, it is equivalent to very efficient data transfer between I/O Die and computing Die through inter-connected technology. . Not only that, but the second-generation AMD EPYC processor is also energy-efficient and very fast, so the requirements for technology are getting higher and higher. The challenge in this regard is that we will continue to pay attention to it in the future.
MC: AMD pays great attention to the construction of the market ecology, and has a lot of partners. What is the ecological strategy of AMD's server partners in China?
Scott Aylor: We have a lot of Chinese partners, and Hongbing and his team are working very hard to support and help Chinese server vendors develop server systems that meet market needs. As a result, our ecosystem is growing and not just multinationals, and local server OEMs in China are developing many server designs.
Liu Hongbing: In fact, our ecological partners in China have two directions. One is the server vendors and local OEMs that people often think of. Today, we still have a lot of global ODMs. Once these ODMs come in, they will become OEMs locally to make server devices. Another direction is that we work with memory vendors, hard drive vendors, and network card vendors. For example, we work with all vendors that support PCIe 4.0 to build an ecosystem.
MC: How does AMD help OEMs and customers get better AI performance through this hardware plus software collaboration optimization?
Scott Aylor: Indeed, AI is now a very promising area, and it spans many industries. You can see that AMD brings a lot of advanced technology and capabilities. We connect the second-generation AMD EPYC processor to a high-performance GPU for the Training Scenario. This configuration supports PCIe 4.0 at the same time, which means that the bandwidth connected to the GPU is doubled, which can accelerate our machine learning and training applications very well. At the press conference, we also showed the Reference Design based on the AMD platform, which is based on two second-generation AMD EPYC processors supporting PCIe 4.0 and eight Radeon Instinct MI60 GPUs, which can greatly improve the performance of machine learning. . In the future, we will also provide this reference design to OEMs.
MC: From the current cooperation between AMD and partners, most of them are two-way cases, and few single-channel cases are seen. What is AMD's opinion and expectations for the single-channel server market?
Scott Aylor: In fact, in terms of single-socket servers, our first-generation EPYC processors are well-marketed, and single-socket servers are well-suited for a wide range of network (Web) technologies, ie those that don't need Enterprise customers with the highest computing performance and one-way server to meet their needs. Not only that, our second-generation AMD EPYC processor server chip has 80 records, a large part of which is recorded from a single-channel server system.
MC: We have seen a significant increase in the number of cores in server processors in the last two years. What is the reason for this trend? What are the key points to make the most of the advantages of the core?
Scott Aylor: One reason for the dramatic increase in core numbers is that Moore's Law is slowing, and it is now more difficult to achieve higher performance for each generation because it makes costs more expensive. In addition, unlike the previous increase in the main frequency, the main way to improve performance is to increase the number of cores. So to increase the number of cores must work closely with the software ecosystem to ensure that their software applications can make good use of all the growing core numbers, so that better hardware and software integration can be achieved. In other words, we can add a lot of cores on the hardware, but at the same time we have to work very closely on the software to let the software fully play the core performance.
