Fastest Computers On The Planet

The latest Green500 list (Excel spreadsheet here) was just released at the end of last month and heterogeneous systems now own the top of the list. The Top 10 systems all use a combination of Intel Xeon (mostly E5) processors paired with NVIDIA K20s. There are now 6 systems listed that have broken the 3,000 MFLOPS/W barrier and TSUBAME-KFC, belonging to the Tokyo Institute of Technology’s GSIC Center, registered over 4,500 MFLOPS/W.

There’s still a ways to go, though, to get to that ExaFLOPS target. Even if we were to allow for a generous 20MW to power this supercomputing beast, that means that we would still need to achieve 1*10(18)FLOPS/2*10(7)W or 50,000 MFLOPS/W. The target seems even farther away when one realizes that the top of the Top500 list, Tianhe-2, belonging to the National Super Computer Center in Guangzhou, comes in at just over 1,900 MFLOPS/W while burning nearly 18 MW of power. Tianhe-2 shows up at #40 on the Green500.

It’s one thing to be at the top of the Green500 list with a machine using almost 28kW, and quite another to scale that efficiency up to the levels of computing to take the top of the Top500. Of particular note here is the Swiss National Supercomputing Centre’s (CSCS) Piz Daint, a Cray XC30 again using Intel’s Xeon E5 processor paired with NVIDIA’s K20x, which not only came in fourth on the Green500 list, but also appears in sixth place on the Top500, the only system to make a top 10 appearance on both the Green500 and Top500 lists.

Table 1. Top 10 of the Green500

At the moment, the top two computers on the latest Top500 list are also the same two heterogeneous systems that appeared in the top positions in June of 2013. IBM’s BlueGene/Q is still holding onto 4 of the top 10 positions on the Top500 list. The announced improvements coming with IBM’s new POWER8 with chips built on IBM’s 22nm FDSOI process surely will be put to good use in terms of making another assault on the lists, but IBM also believes that the heterogeneous trend is serious.

IBM and NVIDIA have announced an agreement to collaborate on building supercomputers that will use IBM POWER8 processors and NVIDIA K40 Teslas. This should lead to a new generation of computers to target the top of these lists. Along with new funding for supercomputing, the charge continues towards reaching the exascale goal and the promise it holds in terms of the new power it would enable researchers and scientists to bring to solving some of the most challenging problems facing the world today.