No ARM in a bit of server market competition

Qualcomm is the latest firm to pitch an ARM-based server concept, but is there a market for them?

Qualcomm has conducted a live demonstration of its upcoming ARM-based server processor, driving home the chipmaker’s intentions to bring high performance ARM servers to market. The firm joins a growing list of vendors that have sought to make a mark with the ARM architecture in the datacentre, yet the industry continues to be dominated by Intel.

Qualcomm first detailed its plans for an x86 server chip back in 2014, stating that it saw an opportunity for such a product thanks to developments in the datacentre market because of the impact of cloud computing. Now, the firm has announced it is sampling its first chips under the Centriq brand and said it expects these to be commercially available in the second half of 2017.

ARM itself has been talking about taking a slice of the server market for even longer, at least as far back as the launch of its 64-bit ARMv8 architecture five years ago. Since then, we have seen a succession of companies line up to back ARM-based server technology, but with relatively little success to show for it so far.

The most recent casualty, according to reports, is Broadcom, which is said to have halted all development work on its ARM-based server chip, dubbed Project Vulcan, following the firm’s acquisition by Avago Technologies.

Meanwhile, AMD is now refocusing server development efforts onto its Zen x86 processor despite kicking off commercial shipments of its Opteron A1100 series of ARM server chips earlier this year. Although AMD is a company with a certain amount of experience in the server market thanks to its ground-breaking original Opteron chips, it has apparently seen little uptake of the Opteron A1100 beyond development systems, such as those from Softiron.


Intel domination

Why should this be? The obvious answer is that Intel has much of the server market sown up with its Xeon server platforms based on the x86 architecture, and many enterprise customers see little reason to switch, especially when they have sunk a great deal of investment into software and tools that run on their existing x86 systems.   

ARM’s proposition has always been that it could take the hugely successful ARM architecture that powers almost every smartphone and consumer tablet in use today, and adapt it to deliver more power-efficient servers.

The continuing growth in demand for compute power from cloud-based services is driving increased energy consumption and costs for datacentre owners, and ARM’s rationale is that switching to less power-hungry systems should help to alleviate this.

But some experts dispute this, saying that in order to ramp up to match the processing power of existing servers, ARM-based systems would lose much of the advantage that the architecture has in energy efficiency.

“Is it so miserly on power that it would make a massive difference in a massive datacentre?  That’s arguable,” says Clive Longbottom, Service Director at analyst firm Quocirca.

“Intel has countered with its own low-power chips – and these have a higher compute power per unit volume and unit costs than the majority of ARM-based systems.  It is only in the likes of a Google, AWS or other proprietary platform environment where the best use could be made of the chip through the use of advanced workload orchestration,” he adds.

These kind of hyper-scale environments were thought to be fertile ground for ARM, because the use of open source software and internally developed tools by these customers would make it easier to port the software environment to the ARM architecture than would be the case with Windows Server applications, for example.

Indeed, Google keeps floating out initiatives regarding its support for alternative compute architectures, including saying that it was working with Qualcomm to design servers based on its ARM chips. But Google is also offering similar support for IBM’s Power processor architecture, leading many in the industry to conclude that the internet giant is simply spreading its bets or trying to gain greater bargaining influence with Intel.


Power takes a different tack

IBM’s approach with the Power architecture is an interesting contrast to ARM, as it contends that its chips are better suited to handling demanding workloads than x86 processors, and also offer better performance per Dollar. In other words, it is offering a higher performance proposition to tempt organisations to switch, rather than a lower power consumption argument.

To bolster this, IBM built technology into its current Power8 chips that makes it easier to integrate accelerator hardware such as GPUs or FPGAs into a system. Dubbed Coherent Accelerator Processor Interface (CAPI), this enables the processor to share its memory space with plug-in offload accelerator hardware, and is being extended in the upcoming Power9 generation.

This ability to handle non-standard workloads was also supposed to be another way that ARM chips could capture a slice of the datacentre market. As an example, HPE developed a cartridge for its Moonshot server system to target transcoding and telephony applications, using chips that integrate ARM CPUs with digital signal processors (DSPs), but this has quietly disappeared from the firm’s website and a HPE spokesperson confirmed that it no longer offers ARM options.

But Qualcomm is not targeting non-standard workloads. The Centriq 2400 series with its 48 processor cores has been designed to tackle the most common datacentre workloads, as was AMD’s Opteron A1100.

“Qualcomm is saying that this is not just a server-focused Snapdragon chip, nor is it aimed at being an offload chip – they do want to take Intel on head-first at the Xeon level.  We will have to see if that is a sensible and workable approach – at the moment, I doubt it,” says Longbottom.

“Qualcomm states that by keeping all the core attributes of the ARM architecture, it should be able to drag many of the ARM developers and community along with it – this smacks of Microsoft and its Universal Apps, something that to date seems to have been a bit of a disaster,” he adds.

So after many years of initiatives and partnerships, mainstream ARM servers still seem as distant a prospect as ever. This is despite support for ARM from all the major Linux distributions, including Suse, Ubuntu and Red Hat, along with ARM ports of tools such as the ubiquitous LAMP stack for web servers.

Perhaps Qualcomm will fare better with its ARM-based servers than the pioneers who came before it, but whether there is really a mass market for such products remains to be seen.