Dell prepares for the arrival of ARM processors in servers (part 3)

Mosaic begins to take shape

In the two preceding parts of the material we examined the history and problems of the issue. In the first part - the glorious way and features of the architecture of ARM processors, and in the second - the current challenges of HPC and the positioning of ARM in this market. At the moment, before the Dell-ready solutions enter the market, we’ll recall some recent news in the final material of the cycle, which now has a clear trend.

ARM

The company, whose name has become a household name, is engaged in the design and development of turnkey solutions based on RISC architecture. It supplies to the market not only the design of processor cores, but also a complete ecosystem for using them for almost any need. It is the largest holder of a license package for RISC processors and everything connected with them.

Currently, the company's flagships are the processors, and more specifically the cores of the Cortex-A57 and Cortex-A53 CPUs. The first is optimized for maximum performance, the second - for maximum energy efficiency. Both kernels are 64-bit, but at the same time they execute 32-bit code in native mode, without recompilation and emulation. They contain a block for floating-point operations and a crypto-core, a memory controller supporting more than 4 GB and a cache memory.

Both are capable of operating in configurations up to 4 cores as a single processor with a shared cache. Such a multi-core chip supports one-to-one virtualization technology.
')
Both support technology big.LITTLE Processing, aimed at improving energy efficiency. With a combination of A57 and A53 cores in one assembly, this technology allows, in case of a need for maximum performance, to use all the processing power of the processor, and in the case of performing background tasks typical of the standby mode, send A57 to a deep sleep and provide small system requirements due to an economical core A53.

Companion cores, in particular, designed for network interaction, and the Mali video processor are available for building ready-made solutions of the SoC class (System on Chip, system on a chip) and FPGA (Field Programmable Gate Array). In addition, for specific tasks, there is an extensive base of fully compatible developments in the form of design of companion cores.

Currently, turnkey solutions that use all of the above technologies are used, for example, in Samsung processors for the flagship phone models of this company.

Altera

The company is just engaged in the development of ready-made solutions for the needs of customers based on the "designers" offered by ARM. FPGA, ASIC and SoC of its development cover a large number of applications of ARM-processors. In particular, they are used in active network equipment, navigation systems, radar systems, control units of missiles and other military facilities.

The company has only recently expressed concerns about the possible return of Intel to the market for RISC processors, but in the end it was a partner for the microelectronic giant in this timely step. It was she who developed the processor core based on the ARM architecture, which will be released by Intel.

Intel

In turn, Intel’s production of RISC processors has long been familiar and mastered very closely. At one time, the whole division of the company was engaged in the development and production of such microcircuits. These developments were initiated by DEC and transferred to Intel in the form of Digital Semiconductor, which later became XScale. Its developed StrongARM line has replaced its own Intel cores, i860 and i960. All these chips were used as device controllers, input-output controllers and network processors, as well as in mobile phones, PDAs, music players, etc. In mid-2006, Marvell Technology Group bought all of Intel's expertise in the field of ARM and RISC processors, including a full license package not only for cores, but also for the entire ecosystem. At the same time, Intel itself also remained the holder of a full license package, but ceased production. Then it looked like “to focus fully on the production of x86 processors”.

Now we see that such an overly narrow focus leads to incomplete loading of Intel's huge production capacity. In addition to the number of factories, it is the developer and owner of the currently thinnest process technology, 14 nm. The fourth generation of Core processors is produced on the basis of 22 nm lithography, but in the fifth generation 14 nm is planned. In general, a finer process technology results in a slight increase in the switching speeds of transistors, a more noticeable decrease in the area of ​​the crystal of the same complexity and performance, and an even more noticeable reduction in power consumption at comparable power.

Despite the fact that the three effects listed above are indisputable and clearly noticeable in theory, in practice they are less visible due to the concomitant increase in the complexity of the crystals, the increase in the cache memory, the inclusion of blocks with new functions, etc. With the same architecture, the advantages of a finer process technology are more clearly visible. In particular, all three highlight the original strengths of the ARM architecture.

For Intel, the contract manufacturing of processors, though not new to it, but the “forgotten” architecture is a good investment of manpower and resources. Having a full package of licenses for ARM, she can enter this market soon with her new developments. Testing a new technical process on a simpler architecture than x86 should cost less and can speed up the release of a new, fifth generation of its own x86 development. In addition, in the context of a shift in the focus of demand, there are persistent rumors that for Intel this is just a test of the pen in contract manufacturing of chips to order by someone else's design, since For a simple factory, this business is not profitable, and for business it is better to work at full load.

So, for the present it is not entirely clear what the new coil of relations between Intel and ARM architecture will turn into, but the first step looks promising.

ARM TechCon

Held at the end of October in Santa Clara, the conference aimed to gather interested parties so that they could demonstrate the seriousness of their intentions and technical readiness for the upcoming mass arrival of ARM architecture processors to the server market. In addition to Dell, Applied Micro / APM, PMC, Fedora and ARM itself took part in it. Conference participants exchanged experiences and practices that are not yet available to an open audience.

To demonstrate a complete installation, the following components were used:

1. ARM showed a working prototype of the Blackbird 3U server based on the 64-bit ARM architecture
2. PMC introduced the 7085H HBA, an adapter for disk drives based on the 64-bit ARM architecture,
3. Dell introduced the PowerVault MD1220 JBOD and Dell Precision T1700 MT disk array with an Intel 10GbE SFP + network adapter as a client,
4. Fedora demonstrated the performance of its server operating system (based on Linux, open source) based on the 64-bit ARM architecture.

In the course of the demonstration, the Gnome shell system monitor demonstrated CPU and network load during streaming video broadcasting, and standard FIO and iperf tests showed CPU utilization during data exchange operations. Thus, a fully functional prototype was assembled.

The next development step for Dell will be the assembly and testing of prototypes of future industrial systems in early 2014, and the delivery of commercial solutions to the market is planned for the second half of the year.

Then we will return to the topic of ARM processors in the mass servers of Dell, the market leader in server engineering and consumer solutions for high-performance computing, HPC, turnkey.