ARM is leading a European Commission-funded datacentre research project to look into using multicore processors to try to deliver energy-efficient cloud-computing services to mobile devices.
Announcing the project, called EuroCloud, on Monday, the chip designer said it will work with Nokia, IMEC (which develops transistors), Swiss academic institution Ecole Polytechnique Federale de Lausanne and the University of Cyprus on the research. EuroCloud has already received an unspecified amount of EC funding, which will last for three years.
The project will use ARM's Cortex-A9 low-power multicore processor alongside 3D DRAM structures and chip packaging to try to make energy savings.
"We need breakthroughs in energy-efficient multicore computing," said Panagiotis Tsarchopoulos, project officer for the European Commission, in a statement. "EuroCloud is an exciting research project that brings leading European low-power technologies from embedded computing to general purpose computing."
ARM said it had been focusing on energy efficiency in mobile devices, but that it now wants to tackle power consumption in the datacentre.
"The performance levels and new functionality supported in our latest processors have reached the level where it is appropriate to see how ARM technology can be applied to solve the massive energy challenges that exist in wireline equipment, including datacentres," said Krisztian Flautner, vice president of research at ARM, in the statement.
Nokia said it is taking part in the project to figure out how it can provide more energy-efficient services using Ovi, the handset maker's mobile application platform.
Read this
Datacentre security: a 10-point checklist
Security is always jockeying for a place at the top of the datacentre agenda, and that's why it makes sense to have a handy checklist of key measures, says Manek Dubash
The project partners outlined their ideas for the design of a power-light server for cloud computing on the EuroCloud website.
"The EuroCloud server concept [is that] emerging cloud-computing applications... favour an increase in thread- and memory-level parallelism and benefit little from instruction-level parallelism. Each client request is serviced by a single or multiple independent threads each running on a dedicated processor core," the project wrote.
"These applications need many simple processor cores with high bandwidth/low latency access to very large memories. Using standard off-chip DRAM is bandwidth-constrained due to limited pin count, slow due to chip crossing and power hungry due to I/O pads and driving circuitry," it added. "To eliminate these inefficiencies and address these issues, we propose to 3D stack DRAM chip on top of the ARM Cortex-A9 processor cores and hardware accelerators eg GPU, video and crypto engines."
The participants predicted that servers based on the design they are developing will be provided to businesses around 2015. The aim of the project is to support hundreds of cores on a single server.
However, one cloud-computing analyst was more cautious on the benefits for businesses brought about by the project. David Bradshaw, research manager for cloud services in Europe for IDC told ZDNet UK on Tuesday: "It's too early days to say that this will deliver specific benefits. But we do need more efficient datacentres."
Bradshaw said low server utilisation was a "big disaster" in the way datacentres are currently run, and that moving server infrastructure to the cloud would vastly increase that utilisation, driving down energy costs.
The research manager said that devices were becoming able to do more of the processing load, and so energy efficiency measures had to involve devices as well as datacentres. Though other chip companies like Intel might be able to learn from the project, Bradshaw said other energy efficiency projects focused on the datacentre would be needed.
