A new breed of ARM64 servers is arriving on the market, offering datacentre managers an alternative to x86 machines...
for high density and high performance computing applications.
While the dominant server platform is still x86-based, there is growing interest in alternative hardware platforms.
But to date, x86-based hardware has been the preferred choice for datacentre deployments, due to low, per unit cost compared to reduced instruction computing (Risc) set hardware.
European figures for the first quarter of 2014 say x86 server revenue increased 3.4%, while Risc/Itanium Unix revenue declined 3.4%, according to analyst Gartner.
Globally, Gartner reported a big decline in non-x86 servers. "x86 servers managed to produce an increase with growth of 1.7% in units for the year and 2.8% in revenue," Jeffrey Hewitt, research vice president at Gartner.
"Risc/Itanium Unix servers fell globally in the first quarter of 2014 – down 19.9% in shipments and a 16.9% decline in vendor revenue compared with the same quarter last year. The 'other' CPU category, which is primarily mainframes, showed a decline of 37.6% year over year in terms of revenue."
More articles on micro servers
But in spite of these poor sales figures, there is growing interest in alternative architectures to x86, because datacentres are big energy users and need to meet Carbon Reduction Commitment regulations.
64-bit ARM powers next-gen servers
HP’s Moonshot was among the first in a new breed of micro servers, based on low-powered processors, aimed at dramatically cutting energy consumption in the datacentre. For instance, the HP ProLiant m700 Server Cartridge is powered by an AMD Opteron X2150 APU running at 1.5GHz, and is designed for desktop virtualisation
In June, AppliedMicro introduced a system on a chip platform based on an ARMv8 64-bit processor, with network and storage offload engines, as well as integrated ethernet.
At the time, a number of server makers announced micro servers based on this chip.
E4 EK003 is a production-ready, low-power 3U, dual-motherboard server appliance with two Tesla K20 GPU accelerators. According to the manufacturer it is designed for seismic, signal and image processing, video analytics, track analysis, web applications and MapReduce processing.
Nvidia is giving its support to OpenPower, presumably in a bid to expand the applications of its high-end Tesla GPUs
Cirrascale RM1905D is a high-density two-in-one 1U server with two Tesla K20 GPU accelerators, which the manufacturer says provides high-performance, low total cost of ownership for private cloud, public cloud, HPC and enterprise applications.
Chipmaker AMD expects its ARM-based Opteron A system on a chip servers will ship in Q4 2014.
All of these servers will require an operating system and development, which is where Linaro, the Linux for ARM system on a chip architectures, comes in.
Power extends reach
IBM has also been looking at the low-powered market.
While sales of IBM’s Power server continue to decline, IBM is working to expand the reach of the chip in the datacentre through a consortium with Google and Nvidia called OpenPower. The Power architecture is IBM’s primary datacentre server architecture, now that it has sold its X Server division to IBM.
As with the ARM system on a chip server architecture, Nvidia is giving its support to OpenPower, presumably in a bid to expand the applications of its high-end Tesla GPUs.
Google’s position is also interesting. In 2012, its CFO, Patrick Pichette said the search engine company was probably one of the world’s largest hardware companies, as it makes its own servers. While there is little detail on the server architecture, Google is built on low-cost commodity servers.
The more servers it runs, the faster search results can be processed, and so there is a direct link between how much it invests in hardware and its revenue stream, which comes from the volume of clicks through to websites.
So, the OpenPower alliance could offer a way for Google to move some of its workloads on to Power-based hardware. Some reports suggest Power can support more virtual machines than x86 and given the Power 8 feature 12 cores and supports up to 96 threads, it is theoretically more capable at running Google search than Intel’s x86 compatible Xeon chips.
A Risc server is still much more expensive than an x86
During IBM’s Q2 2014 earnings call, Martin Schroeter,senior vice president and chief financial officer in IBM's finance and enterprise transformation group, discussed how the company was expanding its newly introduced Power8 server family.
In a transcript of the speech posted on financial website Seeking Alpha, he said: "We expanded our OpenPOWER consortium, doubling the number of alliance members in the second quarter. At the end of June, we had 36 members across 10 countries, including nine in China, so globally diverse. The membership is across the stack, from chip designers to hardware component OEMs, to system vendors to middleware and software providers."
Third-party Power8 server hardware is expected in 2015.
Unless there is signification cost or power advantages, Power8 and ARM-based servers are likely to remain niche products.
Gartner noted that through the alliance IBM wants to engage the Risc/Power platform with Linux closely to create a credible alternative to x86 architectures.
But the analyst stated: "Total cost of ownership in deciding a platform is still the key decision factor in making a purchase, and a Risc server is still much more expensive than an x86. The long-term outlook is still down."
At the end of last year Forrester principal analyst Richard Fichera predicted that if ARM server products could demonstrate significant power-per-workload unit advantages over x86, they would earn a place servicing workloads with extreme efficiency requirements. However, he added: "If ARM can’t deliver better performance than the new generation of x86 plus improved SOC products, ARM will probably fade into limited niches in the server landscape, such as embedded storage controllers and low-power web farms."