Dell is preparing for the arrival of ARM processors in servers
Sometimes important things happen all of a sudden, like, for example, Apple's transition from using proprietary RISC processors to Intel's x86 processors in their desktops and laptops. At one time, the news had the effect of a bomb exploded, and the editors of the online media doubted the expediency of bringing the rumors to the readers to full and official confirmation.
More often, important changes are made up of long undercurrents that finally come to the surface. Everything happens before the eyes of those who are interested in the issue, but sometimes the moment of transition from development to implementation is also unexpected. Dell has been developing in server building for a long time, which is perfectly visible in the wide range of hardware and software that the company offers to its customers. Due to direct interaction with large customers and the delivery of turnkey solutions, Dell is well aware of the current customer needs and challenges that arise in the process of optimizing data center equipment.
One of them is that for a number of tasks compactness and cost-effectiveness of solutions has a higher priority than the computing power of each individual server in the installation. For many years, there have been developments aimed at the widespread use of ARM processors in server hardware as a CPU. As network device controllers, disk arrays, etc. they have been used for a long time, widely and successfully, but that’s how server CPUs are rarely used so far.
')
Let's look at some aspects of the modern development of computer technology that prompted Dell to make such a decision, and we will begin today from afar - what is ARM and where did they come from.
Since the dawn of the computer era, more precisely, with the advent of Intel 8086 processors, in parallel there are two main architectures - x86 and RISC. Their differences begin at the level of instructions executed by the processor, which come from compiled program code, into which written programs turn. Roughly speaking, at the compilation stage it is determined on which processor the program code can be executed. The x86 command set was originally more diverse, to optimize the processing needs of various applications, but the RISC processors handle a limited set of instructions, which imposes some restrictions on their application, but at the same time allows them to more effectively cope with the tasks for which they are intended. The key difference of RISC, which is hard to beat by adapting the program code to the instruction set, is in the absence of floating point calculations. It is critical for multimedia processing and scientific tasks.
Each of the branches of processor building went its own way, and the concepts originally conceived evolved, initially moving away from each other more and more. From generation to generation, the x86 command system was complemented by new extensions. First, the math co-processor in the 386 generation was integrated into the main processor, the Pentium was equipped with a streaming pipeline for processing MMX multimedia, and so on, along the path of complicating both the processors and the code they execute. The processors were getting more and more powerful, but they also required more and more complex “strapping” in the form of a chipset and expansion cards to solve specific problems.
Over time, out of a seemingly frivolous toy, the function of a video card to speed up the calculation of 3D graphics, most in demand in computer games, the new processor architecture has grown. Video accelerators have become mass-parallel pipelines for floating-point calculations. Now they quite successfully cope with cryptographic and scientific tasks and have taken their place in equipping servers and powerful workstations. Tesla from nVidia, FireStream from AMD and Intel Xeon Phi - that's what Quake 2 and Unreal have grown to be.
Both x86 and video accelerators have never sought to save electricity, to be small in size, to be the finished product to perform a particular task. They were and remain powerful, big, hot and universal solutions, seeking to expand the horizons and push the boundaries.
And what about RISC? In order not to go deep into the jungle of its architecture, let's move on to the narrower term ARM. ARM Holdings is the most famous processor developer today with a set of RISC commands. It develops the architecture and the ecosystem for the application of its processors so successfully that today most of the RISC processors are based on its decisions, and the name of the company has superseded the name of the architecture and has become common.
ARM processors have always been relatively simple, small and energy efficient. They were used for highly specialized, objectively set tasks for them. It all started with microcontrollers, they are also called that. They managed network devices, printers, cars on the radio, ATMs, telephone exchanges, radar equipment, GPS ... The list can be continued to an impressive volume. The same core, slightly configured differently and with slight variations on the external binding, could perform a sufficiently large number of applied tasks from this list and be installed in completely different devices. Undoubtedly, over time, these initially simple and mono-functional processors became more complex and were taught new, diverse functions. The key role here was played by the PDAs, over time, due to the convergence with telephones, mutated into personal communicators, exceeding the power of desktop computers five years ago. So, for the most part, this power is provided by ARM processors.
How did it happen that the divergence of the two branches of the development of processor architecture eventually led to convergence? It’s all about the scalability of the ARM architecture and its initial strength - energy efficiency. Since energy storage technologies are developing not as fast as everyone would like, for a handheld, autonomous device like ten years ago, and now it is extremely important to use the limited amount of power stored in the battery with maximum efficiency. Focusing on this priority, the ARM developers, however, constantly and steadily increased their performance.
From the growing power and solving the ever more diverse tasks of phones, a new class of devices was born - tablets, also mostly based on ARM. On the other hand, a long-time competitor moved towards - x86 processors. Their development paradigm was, at first glance, the opposite. Constantly and steadily increasing their computing power, they gradually learned to save electricity. The peak of squandering fell on the Pentium IV family, all further developments have already been released to meet energy efficiency requirements. This was due not least to the fact that an increasing percentage of personal computers began to make laptops, and for them the battery life is a very significant characteristic. Then came the subnotebooks, even smaller netbooks and ... there was a meeting with tablets that grew out of phones!
Summarizing, we can say that the varieties of electronic devices and components that use ARM are much more, and the processors themselves are much more, both in varieties and in absolute terms. But the top step of evolution is still occupied by x86 architecture. They are installed in servers, productive workstations, ultrabooks and laptops.
You can't just wake up like this Monday morning and start producing processors. Command sets, architecture details and technical process are protected by licenses, developer copyright. Since there are three x86 developers, and they have been working in this market for quite a long time, in the process of cooperation and competition between them, cross-licensing takes place, i.e. purchase and exchange of new technologies. It so happened that the full package of documents for the x86 architecture is owned by their developer Intel, the developer of 64bit AMD technology and the company VIA. Enter from scratch in this market now is almost impossible.
Licenses for the recruitment of teams, the architecture and the ecosystem of ARM are owned by their developer ARM Holdings, but for a long time it has been selling the right to manufacture ARM processors and all that is connected with them to third parties for a relatively small amount of money. Many companies produce these processors, both on their own and on a contract basis. Among the largest are Samsung, Texas Instruments and Taiwanese TSMC, which does not develop microchips by itself, but produces them according to the order and design of customers.
It is a factory, i.e. owns production capacities and technical processes and fulfills orders of developers. Developers can produce chips themselves, and they may not even have production at all, such as nVidia.
It is worth noting that Intel is not only the largest developer and holder of various patents, but also the owner of the largest network of factories for the production of chips.
X86 processors were initially running under DOS and UNIX. Microsoft replaced DOS with a large Windows family, and free UNIX spawned many clones, the most viable of which turned out to be Linux, which also spread across many channels, covering more and more tasks and applications. And somewhere close to him - MacOS, who worked in his youth on RISC-processors.
ARM initially worked under the management of its own programs, which were not fully operating systems, in the usual sense. In the future, in the course of complicating tasks, a more powerful and convenient user-interface interface was needed. There is a number of them, but the current trend is that most likely the manufacturer will prefer any of the varieties of Linux, for example, Android OS. But Microsoft, look, drops into the last car of a fast train with Windows RT !!!
As mentioned earlier, ARM processors are produced by many and in huge quantities. Competitive pressure is high, prices for a single basic product are small, income is also generally not very large.
The x86 processor market from monopolization Intel rescues mainly AMD, VIA shares can be neglected. The cost of a unit of production is high, the circulation is rather limited, the profits of producers are substantial.
Speaking about the positioning of the architecture, it can be clearly stated that at the moment most of the areas of application overlap. More or less successfully, with more or less common, but both of them can perform a wide range of tasks, both traditional and newly emerging. So, the Atom processor series is designed by Intel to compete with ARM in areas where they are traditionally strong. A new development of ARM are positioned to work in the composition of high-performance servers.
In the next article, we will briefly review some trends in the development of modern server hardware for data centers and try to understand why ARM is the appropriate answer to modern challenges.
UPD:
Part 2
Part 3
More often, important changes are made up of long undercurrents that finally come to the surface. Everything happens before the eyes of those who are interested in the issue, but sometimes the moment of transition from development to implementation is also unexpected. Dell has been developing in server building for a long time, which is perfectly visible in the wide range of hardware and software that the company offers to its customers. Due to direct interaction with large customers and the delivery of turnkey solutions, Dell is well aware of the current customer needs and challenges that arise in the process of optimizing data center equipment.
One of them is that for a number of tasks compactness and cost-effectiveness of solutions has a higher priority than the computing power of each individual server in the installation. For many years, there have been developments aimed at the widespread use of ARM processors in server hardware as a CPU. As network device controllers, disk arrays, etc. they have been used for a long time, widely and successfully, but that’s how server CPUs are rarely used so far.
')
Let's look at some aspects of the modern development of computer technology that prompted Dell to make such a decision, and we will begin today from afar - what is ARM and where did they come from.
Processor Architecture - A Little History
Since the dawn of the computer era, more precisely, with the advent of Intel 8086 processors, in parallel there are two main architectures - x86 and RISC. Their differences begin at the level of instructions executed by the processor, which come from compiled program code, into which written programs turn. Roughly speaking, at the compilation stage it is determined on which processor the program code can be executed. The x86 command set was originally more diverse, to optimize the processing needs of various applications, but the RISC processors handle a limited set of instructions, which imposes some restrictions on their application, but at the same time allows them to more effectively cope with the tasks for which they are intended. The key difference of RISC, which is hard to beat by adapting the program code to the instruction set, is in the absence of floating point calculations. It is critical for multimedia processing and scientific tasks.
Each of the branches of processor building went its own way, and the concepts originally conceived evolved, initially moving away from each other more and more. From generation to generation, the x86 command system was complemented by new extensions. First, the math co-processor in the 386 generation was integrated into the main processor, the Pentium was equipped with a streaming pipeline for processing MMX multimedia, and so on, along the path of complicating both the processors and the code they execute. The processors were getting more and more powerful, but they also required more and more complex “strapping” in the form of a chipset and expansion cards to solve specific problems.
Over time, out of a seemingly frivolous toy, the function of a video card to speed up the calculation of 3D graphics, most in demand in computer games, the new processor architecture has grown. Video accelerators have become mass-parallel pipelines for floating-point calculations. Now they quite successfully cope with cryptographic and scientific tasks and have taken their place in equipping servers and powerful workstations. Tesla from nVidia, FireStream from AMD and Intel Xeon Phi - that's what Quake 2 and Unreal have grown to be.
Both x86 and video accelerators have never sought to save electricity, to be small in size, to be the finished product to perform a particular task. They were and remain powerful, big, hot and universal solutions, seeking to expand the horizons and push the boundaries.
And what about RISC? In order not to go deep into the jungle of its architecture, let's move on to the narrower term ARM. ARM Holdings is the most famous processor developer today with a set of RISC commands. It develops the architecture and the ecosystem for the application of its processors so successfully that today most of the RISC processors are based on its decisions, and the name of the company has superseded the name of the architecture and has become common.
ARM processors have always been relatively simple, small and energy efficient. They were used for highly specialized, objectively set tasks for them. It all started with microcontrollers, they are also called that. They managed network devices, printers, cars on the radio, ATMs, telephone exchanges, radar equipment, GPS ... The list can be continued to an impressive volume. The same core, slightly configured differently and with slight variations on the external binding, could perform a sufficiently large number of applied tasks from this list and be installed in completely different devices. Undoubtedly, over time, these initially simple and mono-functional processors became more complex and were taught new, diverse functions. The key role here was played by the PDAs, over time, due to the convergence with telephones, mutated into personal communicators, exceeding the power of desktop computers five years ago. So, for the most part, this power is provided by ARM processors.
How did it happen that the divergence of the two branches of the development of processor architecture eventually led to convergence? It’s all about the scalability of the ARM architecture and its initial strength - energy efficiency. Since energy storage technologies are developing not as fast as everyone would like, for a handheld, autonomous device like ten years ago, and now it is extremely important to use the limited amount of power stored in the battery with maximum efficiency. Focusing on this priority, the ARM developers, however, constantly and steadily increased their performance.
From the growing power and solving the ever more diverse tasks of phones, a new class of devices was born - tablets, also mostly based on ARM. On the other hand, a long-time competitor moved towards - x86 processors. Their development paradigm was, at first glance, the opposite. Constantly and steadily increasing their computing power, they gradually learned to save electricity. The peak of squandering fell on the Pentium IV family, all further developments have already been released to meet energy efficiency requirements. This was due not least to the fact that an increasing percentage of personal computers began to make laptops, and for them the battery life is a very significant characteristic. Then came the subnotebooks, even smaller netbooks and ... there was a meeting with tablets that grew out of phones!
Summarizing, we can say that the varieties of electronic devices and components that use ARM are much more, and the processors themselves are much more, both in varieties and in absolute terms. But the top step of evolution is still occupied by x86 architecture. They are installed in servers, productive workstations, ultrabooks and laptops.
The main differences between the two main directions of development of processors now
Licensing and production
You can't just wake up like this Monday morning and start producing processors. Command sets, architecture details and technical process are protected by licenses, developer copyright. Since there are three x86 developers, and they have been working in this market for quite a long time, in the process of cooperation and competition between them, cross-licensing takes place, i.e. purchase and exchange of new technologies. It so happened that the full package of documents for the x86 architecture is owned by their developer Intel, the developer of 64bit AMD technology and the company VIA. Enter from scratch in this market now is almost impossible.
Licenses for the recruitment of teams, the architecture and the ecosystem of ARM are owned by their developer ARM Holdings, but for a long time it has been selling the right to manufacture ARM processors and all that is connected with them to third parties for a relatively small amount of money. Many companies produce these processors, both on their own and on a contract basis. Among the largest are Samsung, Texas Instruments and Taiwanese TSMC, which does not develop microchips by itself, but produces them according to the order and design of customers.
It is a factory, i.e. owns production capacities and technical processes and fulfills orders of developers. Developers can produce chips themselves, and they may not even have production at all, such as nVidia.
It is worth noting that Intel is not only the largest developer and holder of various patents, but also the owner of the largest network of factories for the production of chips.
OS support
X86 processors were initially running under DOS and UNIX. Microsoft replaced DOS with a large Windows family, and free UNIX spawned many clones, the most viable of which turned out to be Linux, which also spread across many channels, covering more and more tasks and applications. And somewhere close to him - MacOS, who worked in his youth on RISC-processors.
ARM initially worked under the management of its own programs, which were not fully operating systems, in the usual sense. In the future, in the course of complicating tasks, a more powerful and convenient user-interface interface was needed. There is a number of them, but the current trend is that most likely the manufacturer will prefer any of the varieties of Linux, for example, Android OS. But Microsoft, look, drops into the last car of a fast train with Windows RT !!!
Market
As mentioned earlier, ARM processors are produced by many and in huge quantities. Competitive pressure is high, prices for a single basic product are small, income is also generally not very large.
The x86 processor market from monopolization Intel rescues mainly AMD, VIA shares can be neglected. The cost of a unit of production is high, the circulation is rather limited, the profits of producers are substantial.
Positioning
Speaking about the positioning of the architecture, it can be clearly stated that at the moment most of the areas of application overlap. More or less successfully, with more or less common, but both of them can perform a wide range of tasks, both traditional and newly emerging. So, the Atom processor series is designed by Intel to compete with ARM in areas where they are traditionally strong. A new development of ARM are positioned to work in the composition of high-performance servers.
In the next article, we will briefly review some trends in the development of modern server hardware for data centers and try to understand why ARM is the appropriate answer to modern challenges.
UPD:
Part 2
Part 3
Source: https://habr.com/ru/post/227213/
All Articles