USA — Gaming Console Supercomputer Can Read, Correct Input

WASHINGTON, Dec. 3, 2010 — Video games have advanced by leaps and bounds in the past few years. What once was a big black box with a bad video ver­sion of ping-pong is now a sleek, motion-cap­tur­ing, high-res­o­lu­tion com­put­er sys­tem capa­ble of net­work­ing around the world.
Mark Bar­nell, direc­tor of high-per­for­mance com­put­ing and the Con­dor Clus­ter project at the Air Force Research Lab­o­ra­to­ry, has used that tech­nol­o­gy to cre­ate a new super­com­put­er.

The Con­dor Clus­ter, a het­ero­ge­neous super­com­put­er built from off-the-shelf com­mer­cial com­po­nents — includ­ing 1,716 Sony PlaySta­tion 3 game con­soles — could change the super­com­put­ing land­scape, Bar­nell said yes­ter­day in a “DOD Live” blog­gers round­table.

The sys­tem com­putes oper­a­tions by the tril­lions per sec­ond –- called “ter­aflops” in the com­put­ing world. Some super­com­put­ers can oper­ate at a quadrillion cal­cu­la­tions per sec­ond, or a “petaflop.”

Bar­nell said the Con­dor Clus­ter also rep­re­sents new ways for super­com­put­ers to increase com­pu­ta­tion­al resources while using less ener­gy. Bar­nell said it’s cur­rent­ly the sev­enth-green­est com­put­er in the world.

“This par­tic­u­lar sys­tem is about half a petaflop, or capa­ble of about 500 tril­lion cal­cu­la­tions per sec­ond,” he said. “In the cur­rent time that we can mea­sure it, it’s about the 35th- or 36th-fastest com­put­er in the world, and with some things that are going to be chang­ing in the next eight or nine months with some upgrades, we could boost it to maybe the 20th-fastest com­put­er in the world, and at the same time make it, at that moment in time, the green­est com­put­er.”

The Con­dor Clus­ter isn’t designed to com­pete with the world’s largest super­com­put­ers, he said. The Con­dor Clus­ter, which cost $2 mil­lion to build, is made for more spe­cif­ic tasks. The cheap­est com­pa­ra­ble super­com­put­ers would cost $50 mil­lion to $80 mil­lion, he said. The high­ly advanced Cray super­com­put­ers are in the $100 mil­lion range.

“So from a price per­for­mance, we’d prob­a­bly beat all of them, but the biggest thing for us was the par­tic­u­lar appli­ca­tions and the hard­ware we chose to build this com­put­er with pur­pose­ly match­es those appli­ca­tions well,” Bar­nell said. “Some of the sys­tems that you might refer to in the top 10 in the world are more of a gen­er­al-pur­pose com­put­er and also run appli­ca­tions that we may not. We’re just going to coex­ist and do some things that we need to get done with this par­tic­u­lar super­com­put­er.”

One area the Con­dor Clus­ter is being used in is neu­ro­mor­phic com­put­ing, or “com­pu­ta­tion­al intel­li­gence.” Essen­tial­ly, pro­gram­mers write algo­rithms to “teach” the com­put­er how to read sym­bols, let­ters, words and sen­tences. By pro­gram­ming the com­put­er to read, in the­o­ry it can be taught to fill in gaps and “think” on its own. The idea is that the com­put­er, when tak­ing in mil­lions of lines of data, could fill in gaps or rearrange the pages in case of human error.

The Con­dor Clus­ter can read 20 pages of infor­ma­tion per sec­ond, and even with 20 to 30 per­cent of the char­ac­ters on the page removed, can recov­er all of the sen­tences and words with about 99.9 per­cent accu­ra­cy, Bar­nell said. The dis­cov­er­ies this com­put­er could lead to would change the face of com­put­er sci­ence, he added.

“We have quite a few research and devel­op­ment efforts, work­ing on those kinds of appli­ca­tions to do con­fab­u­la­tion and pre­dic­tion,” he said. “That will open up a vari­ety of areas which could help a lot of oth­er efforts and a lot of the areas in which the Air Force would like to go.”

Source:
U.S. Depart­ment of Defense
Office of the Assis­tant Sec­re­tary of Defense (Pub­lic Affairs)

More news and arti­cles can be found on Face­book and Twit­ter.

Fol­low GlobalDefence.net on Face­book and/or on Twit­ter