SC11: Vendors tackle high-performance hurdles on road to exascale computing

The 2011 Supercomputing conference (SC11) in Seattle last week saw developers and engineers grappling with a variety of new questions and opportunities.

Many of these challenges centre on exascale supercomputing, which will spawn new  hardware and software platforms over the next several years.

At the heart of these developments is Intel, which had a particularly busy week in Seattle. The chip giant used the event to unveil its first Knight's Corner processors. These chips are based on Intel's Many Integrated Cores (MIC) approach, and will function as co-processors to the Xeon in high-performance systems.

Details on the chips remain scarce. The prototypes demonstrated to the press were among the first to be fabricated, and Intel is withholding basic information such as clock speed and the exact number of cores.

The company did, however, provide eye-opening benchmarks. A prototype system achieved a consistent performance level of 1 teraflop. Such a processing load was considered a supercomputing milestone as recently as the late 1990s, when Intel built the ASCI Red cluster.

Intel also touted ease of development, saying that maintaining a common development platform between Xeon and Knight's Corner will offer a major edge over other multi-core and general-purpose computing on graphics processing unit (GPGPU) development platforms.

One company looking to outpace Intel on the road to exascale is Nvidia, which unveiled a supercomputing development platform called OpenACC.

OpenACC is backed by partners such as Cray and Oak Ridge National Laboratory, and seeks to further expand the ability for software developers to take advantage of the multi-threading abilities of graphics processing unit (GPU) chips for scientific calculations.

AMD and Cray also announced a deal to provide a petaflop-scale scientific analysis cluster for the University of Illinois. The Blue Waters contract was originally awarded to IBM in 2008, but lingering development issues prompted Big Blue to give up on the platform earlier this year.

The OpenCL project delivered yet another tool for supercomputing developers with the ratification of the 1.2 release by the Khronos Group.

Described by developers as an important step in the platform's development, OpenCL 1.2 allows separate compiling of various components for the first time. The tool lets developers share and access pre-compiled libraries, which is an important step in the development of OpenCL.