Fujitsu K Computer Takes First Place in TOP500 List of Supercomputers

Remember that 2.6 petaflop monster supercomputer from China, the Tianhe-1A, you know, only the fastest super computer known to man? That’s so 6 months ago and so slow. This week saw a new supercomputer rise to the top spot, the K Computer from Japan with a dizzying 8.16 petaflops under Linpack – and it’s not even complete.

Built by Fujitsu, the K Computer does not make use of GPUs or other acceleration cards, nor is it powered by AMD, Intel or IBM. It currently combines 68,544 SPARC64 VIIIfx CPUs (a chip co-developed by Fujitsu with SUN Microsystems), that’s 8 cores per CPU for a total core count just shy of 550k.

The name, K Computer, comes from the Japanese word ‘Kei’, meaning ten quadrillion (10^16), referring to the 10 petaflop goal of the project. Each of the SPARC64 Venus CPUs are capable of 128 GFlops, so with some rather basic math, we can determine that the system needs another 9,851 CPUs for 78,125, to reach it’s 10 petaflop peak. It’s pure calculative grunt is also matched by it’s power consumption of 9.89 megawatts, which as high as it may be, still makes it one of the most energy efficient supercomputers out there.

So, from 2.6 to 8.16 in 6 months wihout the aid of GPUs, that’s quite the leap in processing. Give it another 6 months and we could be hitting 12 petaflops or more… maybe. I guess we’re not too far away from those Exascale beasts that Intel have been working on, but hey, that’s in 2018; by that time, the K Computer would be the size of a smartphone, playing Angry Birds with a laser projected holographic display… directly into our brains!