Cloud computing drives low-power server chip market staged a good show

[YORK server channel May 20 news cloud computing and other application trends under the influence of data center construction is growing at an unprecedented speed. In a growing data center, energy consumption again becomes a key factor in data center costs, and as a result, low-power server products have drawn the attention of all parties. There are already Dell, HP and many other mainstream server vendors have introduced low-power server products.

In the low-power server industry chain, chip market is undergoing new changes, the current low-power market, the main processor platforms include Intel and ARM. For both companies, Intel entered the low-power server market from top to bottom, while ARM broke into the data center market from bottom to top.

Following the introduction of the Atom S1200 Series data center SoC platform in the fourth quarter of 2012, Intel officially launched the second generation of Atom SoCs for 22-nm applications for data center applications in September 2013. Atom series itself has the advantage of low power consumption, and in 2013 the latest Avoton platform, Intel will be defined in the cold data processing, and provides better performance and higher energy efficiency, while technology and applications All provide innovation.

ARM's operating model is more unique, not to produce their own processors, but to provide the processor design architecture, designed by the partners. For ARM, the more important partners in the field of servers include AMD. AMD introduced code-named Seattle's Opteron A1100 processor in early 2014 and became the first company in the world to offer a 64-bit ARM server processor. Based on the 64-bit ARM Cortex-A57 architecture, Seattle is based on a SoC single-chip design with four or eight cores at frequencies exceeding 2GHz and manufacturing processes of 28nm and is fully qualified with the current 32-bit ARM chips.

AMD's ARM processor is not just a 32-bit architecture upgraded to 64-bit, but a comprehensive upgrade. On the cache side, Seattle shares 1MB L2 cache for every two cores, up to 4MB. All cores share 8MB L3 cache. There is a tremendous amount of innovation in memory, integrating a memory controller capable of supporting DDR3 / DDR4, which means that AMD ARM will support DDR4 memory in the future, which is a significant improvement. Maximum memory capacity of 128GB, an average of 16GB per core, making AMD ARM ready for virtualization.

ARM and x86 are based on two different instruction sets, respectively, both of these instruction sets are booming, and are very successful. x86 is the representative of complex instruction set CISC, and ARM is the representative of reduced instruction set RISC. Even the name of ARM directly indicates its technology: Advanced RISC Machine and advanced RISC machine. The former is more focused on high performance while the latter focuses on the small size and low power consumption. These two instruction sets are applicable to different workloads, such as CISC even more advantageous in the implementation of high-density computing tasks, RISC can gain the upper hand in the implementation of simple and repetitive tasks.

In any case, x86-based servers are much faster and more powerful than ARM-based systems in terms of performance, and ARM is not at all an x86-based rival in terms of performance and manufacturing processes. However, the strength of ARM is not the performance but the efficiency. ARM adopts the RISC pipeline instruction set and is at a disadvantage in completing the comprehensive work. However, it can exert its advantages in some tasks with relatively fixed tasks.

X86 server structure bridge with the expansion of equipment (such as hard drives, memory, etc.) to connect, x86 server structure can be easily expanded, such as increasing the memory, hard disk and so on. ARM architecture server through a dedicated data interface to the CPU and data storage devices to connect, so ARM memory, memory and other performance extensions difficult (usually in the product design has its memory and data storage capacity), so the use of ARM The structure of the system, generally do not consider expansion.

x86 systems have established a common standard for hardware and software development. Almost all x86 hardware platforms have direct access to Microsoft, Linux and virtually all of today's popular tools, so x86 systems have unparalleled compatibility benefits. Almost all ARM systems use Linux operating system, and almost all hardware systems have to build their own system, and other systems are not compatible, which also led to its application software can not be easily ported, which has been a serious constraint on the ARM system Development and application.

In addition, the use of x86 servers not only has a large number of third-party software to choose from, but there are a number of software programming tools to help you get what you are looking for. ARM architecture because of hardware performance constraints, streamlining of the operating system, as well as system compatibility and other issues constraints, resulting in the ARM architecture server system can not have as many as the x86 system programming tools and third-party software to choose and use, ARM programming Most languages ​​use C and JAVA.

From the above we can see ARM either hardware or software simply can not compare with the x86 architecture, but power is the inherent advantage of ARM. Low-power servers require a low-power processor that does not need to provide strong absolute performance, but must have very low absolute power consumption and ultra-high performance per watt, scalability, and very low sales Price, so that users can configure a different number of low-power processors based on demand.

In fact, from another perspective, the so-called short board or the problem is actually a breakthrough, this is the same for Intel's x86 or ARM. Intel has recognized x86 shortcomings in energy consumption, so Intel has a great tool in low power consumption. ARM has also been to improve the ecosystem, so the low-power server chip market is still behind the good play. Only the market fully competitive, consumers have a more adequate choice of opportunities. This is no exception in the enterprise market.