Arm+gpu Constructing Exascale-time supercomputing Imagination

In the past SC11 conference, we have seen a number of products related to arm servers, such as the arm chip upstart Calxeda announced its energy consumption of 5 watts Energycore arm chip, and the company and x86 server giant HP partnership. The ARM architecture has become more and more prominent, but its painfully slow 64-bit transformation is expected to be completed in 2014.

Even compared to the most energy-saving x86 chips, the ARM processor uses much less power, so many academics agree that the ARM processor is the ultimate answer to the problem of large-scale network services infrastructure. However, when it comes to enterprise and High-performance Computing (HPC) handling trivia, ARM doesn't take it lightly. 32-bit version of the system can not handle enough memory, so the arm chip and its instruction set in the server application and the main x86 processor a relatively short.

But by combining arm chips with the GPU's massive digital computing power, it is possible to combine HPC and analysis with two types of applications: very efficient energy consumption and density without compromising any performance. Similar practices have begun to be used to build exascale supercomputers, and the cost of energy consumption has become a huge barrier to building a exascale system.

Mont Blanc plans to be released

At the SC11 conference, Nvidia announced significant progress. First, the company talks about the Mont Blanc project in Europe (Mont-blanc project), which aims to develop a European Exascale-level supercomputer using energy-efficient commercial components and to match the global supercomputer 500.

The Mont Blanc program will rely primarily on Nvidia's Tegra arm processor as a GeForce GPU Accelerator for flow control. The first prototype will be equipped with 256 arm processors and a large number of Nvidia GPU.

The Cuda Arm development tool, which will be released next year, will include a four-core Nvidia Tegra 3 ARM processor and a motherboard supporting the CUDA technology GPU, as well as Gigabit Ethernet, SATA, and USB ports, whose prices are not disclosed with other details.

The combination of arm and GPU (or other specialized accelerators) will be able to meet the x86 giants in the field of scientific and technological computing, the most powerful challenge. Admittedly, there is no arm HPC software system, and although HPC users are often less willing to dabble in new technology, they will be happy to build and optimize their software if the end result is higher performance and lower costs.

If this combination can successfully prove itself in HPC, then it will only be a matter of time before commercialization and start appearing in the Enterprise data center.

Original link: http://www.theregister.co.uk/2011/11/14/arm_gpu_nvidia_supercomputer/

(Responsible editor: The good of the Legacy)