DISQUS

cmechels · 1 year ago

Comments
By Chris Mechels

This piece should be taken as hype, which is what it is. The "sustained" performance will turn out to be LINPAC runs which demonstrate that LINPAC will run fast and continue to be used for such things. As for running "real" LANL weapons codes at a reasonable portion of these peak speeds, don't hold your breath. The last big, and expensive, machine at LANL, called Q, was an absolute, unreliable DOG, and they swept the whole thing under the rug.

Lest you believe that I am just blowing smoke; although I have a long history in "super" computers at LANL, going back to 1976, here's a bit from a LANL weapons code person, Jim Hill, which I found over at "Slashdot" on the web;

"As a software developer who's worked on the Lab's previous ASC machines (Blue Mountain, Q, Lightning) I can say that once the calculation is run to get a machine atop Jack Dongarra's gee-golly list, it's partitioned, segmented, divided, and subjected to such crappy resource management that if I could trade the entire machine for a pair of coupled 8-core Mac Pros I'd do it in a heartbeat."

Dongarra is the guy who built LINPAC, the code I mentioned earlier.

The real problem with this machine is HOW it was acquired. As Domenici's web site brags; it was done with EARMARKS. Domenici, and LANL, bypassed the normal planning and funding process at the DOE and forced this machine on DOE. This is no way to run a railroad.

Why such an endrun? Perhaps because the previous LANL "super" computers were losers, and the DOE was tired of LANL lies and mismanagement. Well, they have done it again, so we look forward to more lies and mismanagement.

One of the least competent managers of super computers, LANL, has, with the help of Domenici picked our pockets once again. We will get the hype, as usual, and the bad news will be buried. The same old game.