The Green 500 most efficient system list is also interesting, although it hasn't been officially posted quite yet: http://www.top500.org/green500/
Apparently the Chinese built a supercomputer named TaihuLight with 49,960 nodes and a custom processor to capture the #1 spot in the Top 500 supercomputer list at 93 Petaflops. The chips are the "SW26010 processor" which apparently includes multiple modules with a total of 260 cores in each processor. I am taking a SWAG that these cores are a form of MIPs derivative since the Chinese have used that architecture in the past, but there is very little public information about the chip architecture or how they perform outside of LinPack. The article itself mentions rumors about the Alpha architecture.
As an aside, the 40,960 node count using Knights Landing chips at about 3TFlops each would turn in a performance number of around 122 Petaflops.
One other interesting quote from the article to remember when looking at LinPack results without other context:
Memory-wise, each node contains 32 GB, adding up to a little over 1.3 PB for the whole machine. While that seems like a lot, it’s not much memory considering the number of cores it must feed. The much smaller 10-petaflop K supercomputer at RIKEN, for example, is outfitted with 1.4 PB of memory, and most of the other large systems on TOP500 list have much better bytes-to-FLOPS ratios than that of TaihuLight. It also relies on the older DDR3 technology, which is slower and more power-hungry than the newer DDR4 memory.
A few other interestly things:
1. The number of machines on the list that use GPU or other custom acceleration parts actually dropped from the last list in 2015.
2. Intel may not be in the #1 system anymore, but Intel's overall share increased slightly to 455 systems (91%).
3. In order to get on the list a system needs at least 285.9 TFlops of Linpack performance, and 958 TFlops to crack the top 100, so the top-100 mark is probably going over 1 Petaflop by the time they release the November list.
