ARM-based servers: a guide

A close-up view of a circuit board

ARM chips have been the mainstay of mobile devices for many years now but ARM with its ARMv8 processor technology promises much for the datacentre. Announced back in October 2011, several different vendors have produced chips with the aim of taking on Intel in the datacentre.

For the channel, there is hope that ARM-based servers nudge customers into buying more hardware for their datacentres and infrastructure to offset the slump in x86 sales and declining profit margins.

ARMv8 Latest news

15/06/2016

ARM vendors set for growth in 2017

ARM vendors will begin to gain market traction with their newest generation of designs in 2017, according to IDC.

According to an IDC report, Worldwide x86 and ARM Server-class Microprocessor 2016–2020 Forecast, combined x86 and ARM server-class microprocessor revenues will log a compound annual growth rate (CAGR) of 2.2 percent from 2015-2020, reaching $15.3bn in 2020.

Another driver of growth is the use of server microprocessors in system categories outside of servers, such as storage systems and networking systems.

The report says ARM processor vendors, such as Applied Micro and Cavium, have garnered notable design wins and partnerships from communications service providers (CSPs) and systems vendors representing a wide spectrum of end customers and workloads.

In late 2016 and through 2017, several processor vendors will release new products with the potential to compete with Intel's mainstream product line, the Xeon E5-26xx series, according to analysts. These products include AMD's x86-based Zen, Qualcomm’s ARM-based Hydra, Applied Micro's ARM-based X-Gene 3, Cavium’s ARM-based ThunderX 2, and possibly Broadcom's ARM-based Vulcan.

30/05/2016

Cavium previews ARMv8-based ThunderX2

Cavium has revealed details of its ARMv8-based ThunderX2 chip at Computex.

The processor uses 14nm architecture in comparison its predecessor’s 28nm and supports up to 54 cores per socket, which Cavium believes will triple performance from its first-gen chip.

Each core can run at 2.4 and 2.8GHz in normal mode and 3GHz in turbo mode. There is also 40K I-cache and a highly associative 64K D-cache.

There are single- and dual socket configurations with each socket connecting to six DDR4 memory controllers and 12 DIMMs.

ARMv8: Overview

ARMv8 was introduced in 2011 and was the first architecture to sport a 64-bit instruction set, putting it on the map for datacentre workloads. The focus is on power efficiency which is far lower than equivalent Inter offerings.

The architecture also has 32-bit compatibility, meaning that applications built for 32-bit should have no trouble running on a 64-bit processor.

As the chip design is licensed, hardware manufacturers can add extra functionality to the core processor to optimise functions such as networking, etc.

The kind of workloads that could run on ARM are typically those that only require low-power, low-heat, low-cooling servers.

It would seem that ARM-based processors have had a five-year gestation with chips from the likes of AMD, AppliedMicro and Cavium only now going into production.

However, at present, if you want an ARM server, choices are still somewhat limited. There are about six or server models available as of November last year, with the majority of them from Taiwanese manufacturers. HP is currently the only major vendor selling ARM systems, while Gigabyte has an ARM server available that is based on Cavium's 48-core ThunderX SOC.

ARMv8: Features

Power consumption

In terms of energy efficiency, the ARMv8 64-bit processor can perform certain operations more quickly than a 32-bit chip so uses less energy and can power down if necessary.

Also, as ARM is based on RISC (Reduced Instruction Set Computing), it doesn’t need the extra transistors and logic gates to perform complex instructions. Fewer transistors, less energy usage.

ARMv8-based processors such as Cavium’s ThunderX2 is built using the 14nm FinFET process, which should bring about greater energy efficiency leading to a significant reduction of power consumption and heat output.

Performance and cores

Performance is somewhat related to power consumption in as far as it’s all about getting the most performance for as little energy usage as possible. If you need computing oomph and don’t care too much about the power you are using, then ARM isn’t probably going to be part of your datacentre strategy.

That said, ARM is best for energy-efficient tasks that can be spread over many cores and be highly paralllelised. Examples of this are processing big data through distributed programs such as Hadoop.

An example of this is the AppliedMicro XGene 3 processor that will have 32 processor cores, running at up to 3GHz, with eight DDR4 memory channels and 42 lanes of PCI Express Gen 3. It will be manufactured using 16nm 3D FinFet transistors. For a typical workload the chip should manage just over 200,000 IOPS (input/output operations per second).

ARMv8: Summary

ARM-based processors could well be the answer for datacentres looking to cut down on energy usage while increasing capacity for those everyday computing tasks.

The technology is compelling not only for energy efficiency but also that the processors can be optimised for different workloads, allowing organisations to deploy infrastructure in a more flexible and highly tuned manner.

Rene Millman

Rene Millman is a freelance writer and broadcaster who covers cybersecurity, AI, IoT, and the cloud. He also works as a contributing analyst at GigaOm and has previously worked as an analyst for Gartner covering the infrastructure market. He has made numerous television appearances to give his views and expertise on technology trends and companies that affect and shape our lives. You can follow Rene Millman on Twitter.

Latest in Components
Nvidia logo on the side of a building
TD Synnex buoyed by UK distribution deal with Nvidia
AMD logo at the 2023 ChinaJoy summit in Shanghai, China
AMD takes the wraps off new Instinct MI300X chips for AI workloads
Graphic of a CPU on a mult-coloured computerised motherboard
CPU architectures: What’s the difference between ARM and x86 and why does it matter?
A component in a circuit board with red effects
Why subsidies alone won’t grow the UK’s semiconductor industry
Semiconductor and circuit board with data flowing
Waning hardware demand to wipe nearly $70bn from chipmaker revenues
Raspberry Pi CEO Eben Upton at an event
Raspberry Pi CEO Eben Upton on surviving the semiconductor crisis
Latest in Feature
A photo of UNSW's Sunswift 7 car pictured in front of Uluru in Australia's Northern Territory.
How UNSW’s Sunswift Racing and Ericsson achieved cross-country connectivity in Australia’s outback
Matt Clifford speaking at Treasury Connect conference in 2023
Who is Matt Clifford?
Open source vulnerabilities concept image showing HTML code on a computer screen.
Open source risks threaten all business users – it’s clear we must get a better understanding of open source software
An abstract CGI image of a large green cuboid being broken in half with yellow, orange, and red cubes to represent ransomware resilience and data encryption.
Building ransomware resilience to avoid paying out
The words "How effective are AI agents?" set against a dark blue background bearing the silhouettes of flowchart rectangles and diamonds to represent the computation and decisions made by AI agents. The words "AI agents" are yellow, while the others are white. The ITPro Podcast logo is in the bottom right-hand corner.
How effective are AI agents?
An illustration showing a mouth with speech bubbles and question marks and a stylized robot alien representing an AI assistant chirping away with symbols and ticks, to represent user annoyance with AI assistants.
On-device AI assistants are meant to be helpful – why do I find them so annoying?