The US Department of Energy's Argonne National Laboratory has won the HPC Challenge award at SC08 in Austin, Texas.
'It is an honor to be recognised as a winner of the HPC Challenge,' said Pete Beckman, director of Argonne's Leadership Computing Facility (ALCF). 'This award proves that energy efficiency and computational power are not mutually exclusive. We can still push performance boundaries and deliver stellar results while using a fraction of the power typically needed for supercomputers.'
Argonne was the clear winner in two of the four categories awarded in the HPC Challenge best performance benchmark competition which were run using 32 racks of Argonne's Blue Gene/P.
Argonne's score of 103 GUPS (Giga Updates per Second) for Global RandomAccess was almost three times faster than last year's winner. Global RandomAccess measures memory performance and stresses traditional system bottlenecks which are directly correlated to application performance.
Argonne also won the Global FFT category, which measures the floating point rate of execution of double precision complex one-dimensional Discrete Fourier Transform (DFT), which is used to efficiently transform one function into another scoring 5080 Gflops.
The HPC Challenge is a suite of tests that examine the performance of high-end architectures using kernels with memory access patterns considered more challenging than those of the High Performance Linpack (HPL) benchmark used in determining the Top500 list and is sponsored by DARPA High Productivity Computing Systems (HPCS) Program and IDC. The goal of the competition is to focus the HPC community's attention on developing a broad set of HPC hardware and HPC software capabilities that are necessary to productively use HPC systems.
'The HPC Challenge provides an important benchmark for accelerating petascale computation for breakthrough science and engineering and will be an important measure as we begin to work towards the exascale,' Beckman added.
The ALCF is home to DOE's Intrepid, a 40-rack IBM Blue Gene/P capable of a peak-performance of 557 Teraflops (put in other terms, 557 trillion calculations per second). The Blue Gene/P features a low-power system-on-a-chip architecture and a scalable communications fabric that enables science applications to spend more time computing and less time moving data between CPUs, both reducing power demands and lowing operating costs.
