guide    

Below is an example of what the Green500 would look like if a selection of the TOP500 were sorted not by Flops (floating point operations per seconds) but by Flops/Watt.

Disclaimer: We do not believe that this is the best way to rank the Green500. The selection of any power-performance metric is controversial. Below, as an example, we use Flops/Watt as our metric in order to demonstrate what one possible instantiation of a Green500 List would be. However, what values do we use for Flops and for Watts? For the former, do we use peak Flops, Linpack Flops (Rmax), or something else such as HPCC Flops? For the latter, do we use average power over a certain time duration or peak power? In light of currently available information, the table below uses Linpack Rmax for Flops and peak power for Watts. It should be noted that we do not believe that peak power should be used for Watts. Obviously, a different choice of metric or the values in the metric will change the rankings in the Green500. We intend to resolve the issue of which metric, and which values to use through a Request For Comments (RFC) that will be made available on this web site in the near future.

 

Green500 Rank Machine TOP500 Rank Linpack Rmax (GFlops) Peak Power (kWatts) MFlops/W
1 BlueGene/L (IBM) 1 280,600 2,500 112.24
2 MareNostrum(IBM) 5 62,630 1,071 58.23
3 Jaguar-Cray XT3 (Cray) 10 43,480 1,331 32.67
4 Columbia (SGI) 4 51,870 3,400 15.26
5 ASC Purple (IBM) 3 75,760 7,600 9.97
6 ASC White (IBM) 90 7,304 2,040 3.58
7 Earth Simulator (NEC) 14 35,860 11,900 3.01
8 ASC Q (HP) 40 13,880 10,200 1.36

 

 

    guide