[Beowulf] China aims for 100 PF

Prentice Bisbal prentice.bisbal at rutgers.edu
Mon Jul 20 07:52:16 PDT 2015


This article in Wired from 2012 hints at supercomputers much more 
powerful than Jaguar at ORNL for cassified NSA use:

> Meanwhile, over in Building 5300, the NSA succeeded in building an 
> even faster supercomputer. “They made a big breakthrough,” says 
> another former senior intelligence official, who helped oversee the 
> program. The NSA’s machine was likely similar to the unclassified 
> Jaguar, but it was much faster out of the gate, modified specifically 
> for cryptanalysis and targeted against one or more specific 
> algorithms, like the AES.

http://www.wired.com/2012/03/ff_nsadatacenter/

Since it's modified specifically for cryptanalysis, it might not meet 
our definition of 'fastest', similar to the DE Shaw Anton system, which 
can only do one thing, but do it extremely fast.

As a footnote, this Wired article came out over a year before Edward 
Snowden's revelations, but covers many of the same NSA activities in his 
initial revelations.

I think the two systems Killian referenced that are on the Top 500 list 
are probably the 'low-hanging fruit' of the US classified systems. Since 
there are much faster known systems in the world, letting the world know 
they exist isn't that big a deal.

Prentice Bisbal
Systems Programmer/Administrator
Office of Instructional and Research Technology
Rutgers University
http://oirt.rutgers.edu

On 07/17/2015 04:00 PM, Kilian Cavalotti wrote:
> On Fri, Jul 17, 2015 at 12:23 PM, Nathan Pimental
> <nathansaint1 at gmail.com> wrote:
>> What are we guessing is the most powerful classified cluster, and what type
>> of cluster do we all think it is? I'd ask where we all think it is, but I
>> think all of us Americans know that...
> Those 2 seem already good candidates, although obviously not secret
> since they're in the list:
> * http://top500.org/system/178519
> * http://top500.org/system/178445
>
> They're #12 and #13, are labelled as "Government", provided by Cray,
> seem to be totally identical, 3+PF, and vectorized-code friendly. What
> else? :)
>
> Cheers,



More information about the Beowulf mailing list