Powering forward
Following the release of the 37th TOP500 supercomputer rankings, just about every company involved is pleased as punch with one result or another. AMD is happily professing its increased presence in the charts, Fujitsu and RIKEN have grabbed first place with their joint-venture K-Computer and Intel is not only boasting its dominance in the benchmarks, but also espousing its readiness to bring about the next generation of supercomputers.
Kicking off with AMD, the company has seen a 15 per cent growth in the number of systems using its processors featuring in the TOP500 chart. Of the top 100 systems, AMD CPUs power around a third, and over half of the AMD-based systems in the chart use the company's latest 8 and 12-core Opteron chips; with the launch of 16-core processors not far off, AMD will no doubt be hoping that performance-craving clients will be looking to those chips for an upgrade in the near future.
Fujitsu and RIKEN aren't helping the AMD camp, as the K Computer is an SPARC-based system (specifically it uses 2GHz Sparc64 VIIIfx eight-core CPUs). The Supercomputer cluster packs in 68,544 CPUS delivering a total 8.16 petaflops of compute power. Despite the huge amount of equipment, the system manages to achieve a 93 per cent computing efficiency ratio - a testament, Fujitsu says, to the work that has gone into designing an efficient system. What's more, the K Computer is still a work in progress, with the final setup expected to push past 10 petaflops - an intimidating target for would-be challengers to aspire to.
Intel, meanwhile, is taking its placement in 77 per cent of the TOP500 system in its stride, and shrugging its shoulders at the 8-plus petaflop performance of the K Computer. While pleased to note that its Xeon CPUs are apparently the choice component for class-leading supercomputers, Intel is more interested in the future. By 2015 Intel expects the leading system in the TOP500 chart to reach 100 petaflops, and believes the exaflop barrier will be broken in 2018; by the end of the decade, if Intel has its way, the fastest supercomputer will have achieved a massive 4 exaflops of compute power - a mind-boggling figure.
Of course, these supercomputers aren't simply built to run benchmarks, they're used in real-world applications, such as weather forecasting, and codebreaking (not to mention numerous publicly undisclosed activities) where the immense computing power is of real use. So although the vast majority of us will never have direct contact with such systems, the indirect benefits should eventually filter down to our level.
