AMD is ramping up to deliver its first ARM-based server processors later in 2014, but both AMD and ARM see these systems being deployed to meet different requirements than the x86 pizza box servers now dominating the market, which means that today's big-name vendors may not get a look in.
The momentum behind ARM-based servers is gathering pace, with the chip design firm releasing its Server Base System Architecture (SBSA) specifications last week to define a minimum level of hardware that software vendors can target to ensure compatibility across ARM server chips and systems from multiple vendors.
Meanwhile, AMD disclosed further details of its Opteron A1100 Series processors based on ARM's 64-bit architecture, samples of which will be available to system vendors in March while production volume is expected to come in the fourth quarter of this year. The company is also making available a development kit based on early silicon for software vendors to start building code on.
The aim of the ARM-based server ecosystem is not to compete against Intel with its powerhouse Xeon chips, but to meet the requirements of cloud and web service providers who are seeing a huge growth in demand from devices such as smartphones and tablets, according to AMD.
"It's not just about adding 100,000 extra users, but about a couple of billion new device users coming online over the next few years in emerging markets," said Suresh Gopalakrishnan, vice president of AMD's server business unit, speaking at an event attended by V3.
These demands impose different challenges on the data centre than traditional application workloads, according to Gopalakrishnan. Specifically, they are relatively lightweight workloads – such as users searching for something – but there are a great many of these workloads running at any given moment.
This favours a larger number of smaller, less powerful servers, according to Gopalakrishnan. The ARM architecture, with its low power consumption, is ideal for the job because it enables such servers to be packed together densely without overheating or overloading the power supply, he added.
Gopalakrishnan showed pictures of prototype server nodes that are "about the size of a smartphone" and consume less than 20W each, he said. However, it still supports 32GB of memory, and with 24 of them fitting inside a single 2U rack-mount enclosure. "It won't look anything like the servers you've seen in the past," he said.
Not only that, but the majority of these systems are more likely to be custom-built for the service provider's exact requirements rather than being manufactured in the traditional way by a major enterprise server vendor, according to ARM. This is already the case for big internet firms such as Google and Facebook, the firm claimed.
"Things need to change, because data centre costs cannot keep escalating at the same rate as the volume of data going through them is," said Ian Drew, chief marketing officer for ARM.
On the face of it, this would seem to be bad news for server vendors such as HP, which started its HP Moonshot project with the notion of high-density ARM-based servers in mind, while rival Dell is also developing its own ARM-based microserver platforms.
However, Ovum principal analyst Roy Illsley said he did not entirely agree with this viewpoint, and that custom-built servers will only ever be niche.
"Some service providers will want to design and build their own, but I am seeing a move towards wanting to buy hardware that is standardised [in order] to use software defined data centre (SDDC) tools as the management and control layer," he said.
Gopalakrishnan conceded that ARM-based systems will have to fit with existing infrastructure and environments if they are to be accepted.
"This has to slide into that kind of environment. If not, there will be big operational expenditure issues," he said.
He also conceded that x86 servers are likely to remain the mainstay in enterprise data centres for some time to come because of the existing software ecosystem, saying that it is very much a "horses for courses" situation.
"You're going to see some situations where x86 works better and some where ARM works better," he said.
Meanwhile, both AMD and ARM are talking up the effort going into ensuring that the broader software ecosystem is ready to support customers looking to deploy ARM-based servers, especially on the open source side.
"We need tools like Java, GCC (the GNU Compiler Collection of developer tools), and the Lamp (Linux, Apache, MySQL, PHP) stack, and with the work we're doing with the Linaro group, all of these things are pretty much ready," Gopalakrishnan said.
Intriguingly, Microsoft was listed as one of the vendors that had contributed input and support on the Server Base System Architecture (SBSA) specifications. This has sparked inevitable speculation about whether Microsoft is planning to support ARM-based systems with some of its enterprise server software, or even the Windows Server operating system itself.
Microsoft has so far denied that it has plans for any such move. In a statement sent to V3, the firm said: "Microsoft is participating in SBSA as part of our ongoing industry collaboration, but we have no information to share about the possibility of an ARM-based Windows Server."
However, as the Windows NT kernel has already been ported to the ARM architecture for Windows Phone, it is clearly not beyond Microsoft to offer this in fairly short order if it saw demand.
ARM's Drew would also not speak about any such plans, saying that "ARM is an IP company, and we leave the product decisions up to our licencees."
However, AMD is optimistic about the future for ARM-based servers, with Gopalakrishnan saying that he expected these to make up 20 percent of all server shipments worldwide in about five years' time.
"This isn't just a one chip event for us, there is a roadmap for ARM server chips beyond the Opteron A1100 Series," he said.
New Vikendi map adds snow, snowmobiles and new aural and visual twists
Faults and bad weather ground SpaceX, Blue Origin, Arianespace and United Alliance
New regulation expected to cut greenhouse gas emissions by about 17 million metric tonnes between 2020 and 2050
Molybdenum ditelluride is a two-dimensional material that can be easily stacked into multiple layers to create a memory cell