A post at Next Big Future got me to thinking about the shape of the known roadmap for the multi-PFLOPS (and larger) machines. I think it would be a fun insideHPC reader project to collectively document all of the extreme scale machines that are on the horizon right now.

I’m specifically not including SETI@home kinds of clusters here; a super has to be a collection of computing resources intended to be used (at least in substantial part) on a single task or relatively small number of tasks (relative to the number of cores in the machine) in order to be considered. So a supercomputer commissioned to run many multi-thousand core weather simulations (for example) would count, but Google’s datacenter doesn’t count. Likewise, botnets don’t count either. Also, to be listed, the machine has to have a planned peak performance greater than 1.999 PFLOPS, and has to be further along that “wouldn’t it be cool if we could build this machine.”

I’ll get things started; post changes or additions in the comments and I’ll update the list.

1. ORNL, 2 PFLOPS (Jaguar), Cray (Opteron/Magny-Cours), 2009
2. RIKEN, 10 PFLOPS supercomputer, Fujitsu (Sparc64 VIIIfx), 2010
3. NCSA, 10 PFLOPS (Blue Waters), IBM (POWER), 2010
4. NASA, 10 PFLOPS (Pleiades), SGI (Altix ICE/Xeon), 2010
5. DOE NNSA, 20 PFLOPS (Sequoia), IBM (Blue Gene), 2011