▲Click above Leifeng Network Follow
Unless there are significant issues with Intel and AMD’s server sales, these non-x86 servers may struggle to become mainstream in this market.
Article by | Bao Yonggang
Leifeng Network notes: As AMD competes with Intel in the server market again, the era of 50% operating profit margins in the server chip market has ended. Additionally, IBM’s Power servers, Arm servers, and the promising RISC-V architecture chips are all trying to carve out a share of the x86 architecture server market. However, unless Intel and AMD face significant sales problems, non-x86 servers may find it hard to achieve great success. Nevertheless, this market will not be boring in the future.
When 2.58 million servers are shipped in a quarter, it feels disappointing as the overall economy slows down, which may signal an early indication of a global economic slowdown. However, the slight decline of 5.1% in server shipments in the first quarter of 2019 may primarily be due to hyperscale cloud service builders investing at the end of 2017 and early 2018, and then waiting for new processors from Intel and AMD.
According to IDC statistics, in the quarter ending in March, server shipments reached 2.58 million units, with revenue growing by 4.4% to $19.8 billion. Moreover, the configurations purchased by server buyers are becoming increasingly diverse, with greater demand for large flash storage; some require GPU or FPGA accelerators, as they do everything possible to maintain revenue growth, even if it may not grow as we expect.
As far as we know, no original equipment manufacturer or original design manufacturer is rolling in black ink, let alone swimming in it. But as the average selling price of servers drops to levels seen in 2000, it needs to rise again after 2010. The following chart is based on IDC data, showing the average selling price of x86 servers since the first quarter of 2009:
This is almost a mirror image of the decline in average selling prices of servers from 1999 to 2009, as proprietary systems and RISC/Unix systems have fallen out of favor, contributing half of system revenue in the mid-1990s. It is important to understand that before x86 processors, Linux, and Windows Server matured, it was the preferred choice for Internet companies.
When we began tracking servers, the Great Recession was underway. Coincidentally, at that time, AMD was exiting the server market due to its Opteron processors, while Intel’s “Nehalem” Xeons borrowed many experiences from Opteron. During the Great Recession, server virtualization was maturing, and Intel began providing 64-bit multi-core processors, leading to rapid revenue growth for its data center group, along with a significant increase in profits due to a lack of competition.
The Great Recession had many impacts on the entire industry, one of which was that it relegated all non-x86 processors to niche status. RISC/Unix vendors could not provide performance that matched Intel Xeon processors running virtualization stacks with Linux or Windows Server. Meanwhile, as online applications and cloud computing became popular, the computing capacity demand from hyperscale and cloud builders also surged by an order of magnitude. Therefore, Intel’s quarterly sales have been continuously rising, even as core counts and instructions per clock cycle are steadily increasing, which is a conventional upgrade for processors.
If we assume that IDC statistics on the average number of cores sold per quarter and the IPC levels of chips can provide an overall performance capacity metric (in our case, computing integer performance, excluding floating-point performance); we can derive server sales revenue from IDC statistics. If you put these two lines together, it illustrates the increase in global server computing capacity over the past decade. Additionally, it shows how price/performance has improved significantly as RISC/Unix machines have disappeared from the market.
It is important to note that the market occasionally sees large purchases of servers followed by reductions, which are related not only to CPU product cycles. Additionally, the decline in computing capacity from the fourth quarter of 2018 to the first quarter of 2019 was the largest decline experienced by the server market in the past decade, but it actually represents a slight normal regression, primarily linear.
In fact, the capacity of servers after the decline is still 4.4 times larger than the capacity of servers sold globally in the first quarter of 2009. Hyperscale and cloud builders exited in the third and fourth quarters of last year, but enterprises are still driving growth in this market. When enterprises, hyperscale, and cloud builders all exit in the first quarter, it naturally leads to a downturn. But more precisely, hyperscale developers and cloud builders made purchases before the statistics were released, and frankly, this belongs to a return to “normal.”
This chart also indicates that the relative price/performance of servers dropped significantly between 2009 and 2011, but the rate of decline slowed afterward, as Intel maintained its dominant position in the data center. Of course, Intel is increasing support for more integers in its products, enhancing memory speeds for balance, flash is replacing disks, and vector engine performance is growing faster, while integer performance costs are roughly the same as three years ago. Overall, this is an order of magnitude cheaper than during the Great Depression.
In the past two quarters, the price/performance curve has begun to bend downwards, as AMD has re-entered the competition, and core counts will rise, prices will drop, reaching a new level. As far as we know, as prices drop, profits will also decline. The question for the future is whether the growth in market demand for AMD’s Rome Epyc processors and Intel’s Ice Lake Xeons can offset the decline in server profits or even enhance profits.
In any case, we believe that the era of 50% operating profit margins in the server chip market has ended unless AMD and Intel reach an agreement to artificially maintain high prices. This is nearly impossible and also illegal. These profits will never return, just as the x86 Xeons did not lower standards to compete with RISC/Unix during the transition from scientific workstations to data centers. Arm should be a competitor, leading price reductions from smartphones to data centers, but that has yet to happen.
Funding is essential to maintain the sustainability of company operations. From IDC statistics, we can see how much revenue OEMs and ODMs generate. We do not know whether any of them are actually profitable, and we suspect that in many cases, the profits from OEMs and ODMs are minimal and may gain more profits through services. Even from the financial performance of publicly listed companies, it is difficult to determine whether they are actually profitable, as the data contains many interfering factors.
Below is the shipment and revenue situation of suppliers in the first quarter of 2019 and the fourth quarter of 2018:
It is worth mentioning that IBM did not make it to the top five suppliers this time, as its System z14 upgrade cycle has gradually passed, even though Power Systems business has seen some growth. Selling the X86 server business to Lenovo has not helped IBM improve profitability. IBM still does not seem to earn as much money in systems as we think, and when competition is fierce, price declines will cause profits to drop faster.
Here is a comparison of major OEM and ODM revenues since 2009:
Dell is the world’s number one system provider, and it has been its most important source of revenue for some time. HP, including its partner H3C in China, ranks second. Dell continues to grow, HPE remains stable, but it is unclear whether these two companies are truly making substantial profits from selling servers.
In the first quarter, Inspur (including its collaboration with IBM on Power Systems in China) and Lenovo and Cisco Systems were almost tied in IDC statistics. Inspur achieved $1.22 billion in sales, a growth of 36.4%, far exceeding the overall market growth; Lenovo’s revenue was $1.14 billion, a growth of 3.9%; Cisco’s revenue was $1.05 billion, a growth of 6.9%. All ODM manufacturers generated $4.55 billion in sales, a decrease of 1%, while the remaining market grew by 2.7% to $4.34 billion.
In 2019, the revenue distribution between x86 and non-x86 servers changed, and it is difficult to remember the time when, during the Great Depression, the sales revenue of x86 and non-x86 server systems was almost the same. As cloud builders and hyperscale personnel began to build their millions of server nodes, this change is accelerating.
The sales of IBM mainframes and Power platforms, a few Arm systems, and other legacy systems (mainly Sparc and Itanium systems) have reached a stable state of about $1.5 billion per quarter, with revenue or expenses of $200 million. The dominance of non-x86 systems is IBM System z mainframes, based on the Power ecosystem, while Arm still constitutes a very small part of the data center.
This may change, but the possibility is decreasing as AMD re-enters the competitive market. Arm chips may eventually enter the networking and other devices of the data center even more than they do now, without truly impacting the server market unless Intel or AMD’s x86 server sales face issues. It will be interesting to see Arm gain an advantage once Arm server chips and software stacks are perfected. However, for now, enterprises are more concerned with how many of their servers might use AMD Epycs rather than Intel Xeons.
RISC-V could also be a favorable competitor in this field, but in any case, this market is certainly not boring. Leifeng Network
Translated by Leifeng Network, via nextplatform
◆ ◆◆
Recommended Reading
Google lobbies US government to lift Huawei Android ban; Sohu joins social battle to launch Fox Friends; Cook: There will be no incident of China banning Apple
Smartwatch eSIM, the first step towards independent personality
Interpretation | What is Apple’s real intention behind launching iPadOS?
10 things 5G can do that 4G cannot
5G commercial licenses will be officially issued; China Post collaborates with Huawei; Apple establishes its first App design and development accelerator in China
CCF-GAIR 2019
Hosted by the China Computer Federation, the 2019 Global Artificial Intelligence and Robotics Summit, co-organized by Leifeng Network and The Chinese University of Hong Kong (Shenzhen), will be held from July 12 to 14, 2019 in Shenzhen.At that time, Nobel Prize winner James J. Heckman, Chinese and foreign academicians, world top conference chairs, and well-known fellows, along with many heavyweight guests will personally attend, to explore the complex survival situation in the fields of artificial intelligence and robotics.
