Sil­i­con Val­ley is power be­hind su­per­com­put­ers

The ad­vanced ma­chines are used to crunch mas­sive amounts of data.

Los Angeles Times - - BUSINESS BRIEFING - By Pete Carey Carey writes for the San Jose Mer­cury News/ McClatchy.

Sil­i­con Val­ley is famed for spawn­ing the desk­top, mo­bile and cloud com­put­ing rev­o­lu­tions. What is less known is that it’s one of the nerve cen­ters for build­ing the world’s fastest num­ber­crunch­ers. Once con­fined to big na­tional lab­o­ra­to­ries, su­per­com­put­ers are now in de­mand to crunch mas­sive amounts of data for in­dus­tries such as oil ex­plo­ration, fi­nance and online sales. The val­ley’s strong hand in that busi­ness was high­lighted in April when In­tel landed the prime con­tract to de­sign a $200-mil­lion su­per­com­puter named Aurora to be housed at the Argonne Na­tional Lab­o­ra­tory in Illi­nois. Aurora, de­vel­oped in part­ner­ship with Cray of Seat­tle, is likely to be­come the world’s fastest su­per­com­puter when it goes online in 2018. With Aurora’s new ar­chi­tec­ture, the Santa Clara chip com­pany ap­pears to be tak­ing aim at a big­ger slice of what will soon be a $15-bil­lion to $20-bil­lion com­mer­cial mar­ket for “high-per­for­mance” com­put­ers that can give a com­pany a com­pet­i­tive edge. “Right now in the oil and gas in­dus­try, there’s an arms race to see who can get the big­gest su­per­com­puter,” said an­a­lyst Steve Con­way of the re­search firm In­ter­na­tional Data Corp. Energy com­pa­nies use su­per­com­put­ers to pin­point oil de­posits. Car com­pa­nies use them to crash vir­tual cars in safety tests. Proc­ter & Gam­ble uses high-per­for­mance com­put­ing to de­sign de­ter­gents and sham­poo and even potato chips.

Among the val­ley’s su­per­com­puter mak­ers or com­po­nent sup­pli­ers:

Sil­i­con Graph­ics In­ter­na­tional in Mil­pi­tas makes sys­tems for na­tional lab­o­ra­to­ries — there’s one at the NASA Ames Re­search Cen­ter in Moun­tain View — and pri­vate in­dus­try.

In­tel al­ready sup­plies the chips for close to 95% of all high-per­for­mance ma­chines, which in­cludes su­per­com­put­ers and their slightly less pow­er­ful cousins, ac­cord­ing to In­ter­sect360, a Sun­ny­vale con­sult­ing firm.

Hewlett-Packard ranks just ahead of IBM as a sup­plier of the world’s fastest su­per­com­put­ers, ac­cord­ing to a Top 500 list main­tained by Berke­ley re­searchers. It had 35% of the rev­enue from high-per­for­mance sys­tems sold last year, ac­cord­ing to In­ter­na­tional Data Corp.

Nvidia in Santa Clara sup­plies ac­cel­er­a­tor chips — some­thing like the tur­bocharger in a car — used in a grow­ing num­ber of su­per­com­put­ers. Its “Tesla” pro­ces­sors are used in ma­chines at Google, Face­book and Baidu for speech, video and im­age recog­ni­tion.

AMD in Sun­ny­vale makes the Opteron pro­ces­sor that pow­ers the Ti­tan su­per­com­puter at Oak Ridge Na­tional Lab­o­ra­tory in Ten­nessee and is used in Cray’s pow­er­ful XE6 su­per-


8 Mel­lanox Tech­nolo­gies in Sun­ny­vale makes leadingedge net­work­ing gear for high-per­for­mance com­put­ers and is com­bin­ing with IBM and Nvidia to build two new su­per­com­put­ers for the U.S Depart­ment of Energy’s labs at Oak Ridge and Liver­more, Calif.

Think of a su­per­com­puter as a clus­ter of tens of thou­sands of Mac work­sta­tions per­form­ing to­gether like an or­ches­tra to process bil­lions and tril­lions of bits of data ev­ery sec­ond, some­times for hun­dreds of users.

Su­per­com­puter prices run from $500,000 to more than $100 mil­lion. Some are gen­eral-pur­pose ma­chines that can per­form tasks like 3-D mod­el­ing while host­ing large num­bers of users at the same time. Oth­ers are used for one task, such as run­ning a cloud-based ser­vice.

“There used to be a few hun­dred su­per­com­put­ers sold in the world each year be­cause the prices were so high — $10 mil­lion and up,” said an­a­lyst Con­way of IDC. But prices have fallen so sharply for pow­er­ful ma­chines that “these days, com­pa­nies and small or­ga­ni­za­tions that wouldn’t think of get­ting one be­fore can do so.”

Justin Sul­li­van

IN­TEL SUP­PLIES the chips for close to 95% of all high-per­for­mance ma­chines.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.