GOVERNMENT

On May 23rd Supercomputing Online received the following letter to the editor via our site’s “Submissions” feature. The letter addresses our May 20th coverage of the dedication of Los Alamos National Lab’s Metropolis Center for Modeling and Simulation and specifically the new “Q” system which is presently in part, housed in that center. The letter was submitted anonymously. We leave it to you, our readers, to form your own opinion on it. As always, we welcome your comments. ---- An excerpt: “This note is in regard to your May 20th piece on the new LANL supercomputer center. It seemed that the real news was totally missed, namely that LANL's ASCI "Q" is in trouble.”

---------- Steve, This note is in regard to your May 20th piece on the new LANL supercomputer center. It seemed that the real news was totally missed, namely that LANL's ASCI "Q" is in trouble. The word out here is that LANL and Compaq have been renegotiating the contract to reflect reality. The contract calls for Q to run at 30 Tflops and be delivered "by 2002". The recent IDC benchmark of "Phase 2" shows the first 1/3 of the machine as 4096 processors running at 1Ghz. This translates to 8.192 Tflops, not the 10 Tflops required. From the Compaq website we learn that the next "Alpha" release is the EV7, a chip which is big, hot, and with no clock speed improvement. The smaller, cooler, faster EV79 is not due until 2004. All this seems to mean that Compaq could not deliver the Q that LANL ordered. Notably, LANL and Compaq are being very coy about numbers of processor and clock rates, and these numbers are not mentioned in coverage of the their new center. So, it seems that Q is a bust, which is being covered up. Even worse, the Alpha chip has no future as Compaq ended that endeavor, as announced last year, by committing to Itanium for the future. This failure, concealed from the taxpayers and, it seems, even the Dept of Energy, comes on top of LANL's failure with "Blue Mountain" 3 Tflop machine. This machine, from SGI, had serious reliability problems, concealed by LANL/SGI. As a result it was not useful for the nuclear simulations required for the stockpile stewardship mission. The first "milestone" of SS was run by LANL on the Sandia "Red" machine. The recent 3-D milestone was run by LANL on the LLNL "White" machine. LANL marveled at how great the phone connections were, "like being there". Cover-up for the busted Blue Mountain. After wasting over $90 million on Blue Mountain, it seems that LANL has just wasted over $200 million on Q. This ~$300 million waste of the taxpayers money is a huge cover-up with no coverage by the media. That is the real LANL story.