<html><body>
<DIV> </DIV>
<BLOCKQUOTE style="PADDING-LEFT: 5px; MARGIN-LEFT: 5px; BORDER-LEFT: #1010ff 2px solid">
<P>-------------- Original message -------------- <BR>From: Lawrence Stewart <larry.stewart@sicortex.com> <BR><BR>> <BR>> On May 20, 2008, at 8:39 AM, andrew holway wrote: <BR>> <BR>> > Okay, for those unwilling to leap the mental chasm :) <BR>> > <BR>> > Would anyone care to give me what they believe to be the definition of <BR>> > Capacity / Capability Computing. <BR>> <BR>> In the HPC context, these words are used by the big labs to refer to <BR>> different types <BR>> of acquisitions. The largest systems (Roadrunner, any Petaflop system) <BR>> are <BR>> "capability" systems, the idea being that they can do something that <BR>> really <BR>> can't be done with a lesser machine. <BR>> <BR>> "Capacity" machines tend to be run-of-the mill blocks of 10 TF or so, <BR>> 1000 <BR>> cores or so, that bulk up the aggregate throughput capacity of the <BR>> collection of systems. They are intended for large!
number
s of medium <BR>> size jobs. <BR>> <BR>> So Sandia's "Thunderbird" seems to be viewed as a capacity system, <BR>> nothing <BR>> terribly special about it, but it can get a bunch of work done. <BR>> <BR></P>
<P>As terms of art in HPC, I think Larry has it right. One might add that "capability" systems would be expected to include some leading edge non-standards-based technology to provide the extra capability (processor type, interconnect, storage, etc). So in the case of RoadRunner that would be the integrated IBM Cell processors. On the other hand, running a first-of-its-kind problem on a "capacity" system is possible and may lend it some capability cachet. The type of problems run on the system should play into the definition. </P>
<P>Regards,</P>
<P>rbw</P></BLOCKQUOTE></body><br />--
<br />This message has been scanned for viruses and
<br />dangerous content by
<a href="http://www.mailscanner.info/"><b>MailScanner</b></a>, and is
<br />believed to be clean.
</html>