Intel Ships New Six-core Core I7 Chip, Cuts Chip Prices

PC World

 

Intel on Monday introduced a six-core Core i7 desktop processor targeted at enthusiasts like gamers, while also cutting prices of some desktop and server chips by up to 48 percent.

The company announced the Core i7-970 processor, which will run at 3.2GHz and have 12MB of L3 cache. This is the second six-core desktop chip launched by Intel after the Core i7-980X Extreme Edition, which was launched in March.

Intel offers standard Core i7 processors for high-end desktops and Core i7 Extreme Edition processors, which are expensive chips targeted at specific users like hardcore gamers who may be looking for extreme performance. The Core i7-970 is an attempt to trickle down six-core chips and fast performance from the Extreme Edition to the main Core i7 lineup, an Intel spokesman said.

The new Core i7-970 should bring a new level of performance to Core i7 lineup. Intel's 980X Extreme Edition processor has captured many benchmark crowns, outperforming even the company's Xeon server chips.

The Core i7-970 is priced at US$885 in units of 1,000. The processor is made using the 32-nanometer manufacturing process, which helps boost system performance compared to chips based on earlier architectures.

The company also cut the price of the lower-end Core i7-870 processor by 48 percent from $582 to $294, according to the company's price list (PDF), which was published on July 18.

The price was cut to put the offering in line with the company's other Core i7 offerings, the spokesman said. Intel also cut a numerous dual- and quad-core Xeon server and workstation chips from its offerings, including the older 5400 series of chips belonging to the X, L and E series. The company also cut the price of the Xeon X3470 processor by 44 percent from $589 to $328.

Intel on Monday also started shipping the new quad-core Core i5-760 desktop processor, which will run at 2.8GHz and includes 4MB of L3 cache. The chip is made using the company's older 45-nanometer manufacturing process.

Spending Soars on Internet's Plumbing

The Wall Street Journal

Behind the recovery in business spending is a surge in purchases of the computers that form the backbone of the Internet, as companies scramble to meet growing demand for video and other Web-based services.

The need to reach customers and employees over the Web is driving furious demand for server systems, the machines that power corporate computer rooms.

Many companies are stocking up on new Dell servers, which typically cost a few thousand dollars apiece, to replace older machines with more energy efficient models or systems with more powerful processors.

Also, an increasing number of businesses are turning to outsourcing companies, which manage computer rooms for customers and in many cases are sharply stepping up purchases of servers to keep up with rising demand.

"We've been buying thousands of computers this year," says Doug Erwin, chief executive of ThePlanet.com Internet Services Inc., a Houston-based company that runs data centers to offer computing services. ThePlanet says it now owns about 50,000 Dell Inc. servers.

International Business Machines Corp., one of the biggest vendors of servers, said Tuesday that sales of industry-standard servers and IT services jumped 30% in the second quarter, after rising 36% in the first quarter.

The buying activity became apparent last week, when Intel Corp. said quarterly revenue from its unit selling server chips rose 42% from a year earlier, while shipments driven by Internet-related companies' purchases nearly tripled.

Growth in Web traffic isn't a new phenomenon, but computer purchasing to keep up with demand is accelerating because of improving economic conditions and technology that makes purchases of new computers pay off more quickly.

On Thursday, Internet giant Google Inc. reported $476 million in capital spending, including spending on servers and other hardware. That was more than triple the amount it spent a year earlier.

Unlike Google, many companies are side-stepping the costs of building their own computer rooms, opting to place servers they buy in "co-location" centers that maintain machines and offer Internet connections.

Rackspace Hosting Inc., a San Antonio, Texas, company that runs data centers, says it added 9,152 servers in 2009, plus about 3,000 more in the first quarter of this year. Savvis Inc., a competitor based in Town and Country, Mo., says it has purchased more than 80% more servers over the last 12 months.

"All I see all day is trucks coming up to our loading docks dropping off servers," says George Slessman, chief executive of i/o Data Centers LLC, a Phoenix-based company. He says the number of customers that have installed servers in its computer rooms has risen from 140 at the beginning of 2009 to nearly 400 now.

The market research firm IDC puts spending on cloud-computing, a term that includes delivering computing capacity over the Internet, at $16.5 billion in 2009, and projects spending in the field will increase 27% a year through 2014—with the number of servers deployed in cloud applications expected to triple to 1.35 million over that period.

Forrest Norrod, Dell's vice president and general manager of server platforms, says the company has seen "triple-digit increases" in its cloud-related business year over year. "The cloud side is growing faster than the rest" of the server market, Mr. Norrod says.

There are several reasons. Companies keep stepping up the use of the Web to reach customers and adding features like video streams that require more computing power and faster network connections.

Such operations generate huge volumes of data, which have forced companies to buy more-powerful servers to help analyze the information, says Mike Long, chief executive of Arrow Electronics Inc., which sells servers and distributes chips and other components.

Meanwhile, companies that stocked up on servers over the past decade have struggled to find space, electrical power, colocation in Maryland, and labor to keep them running. Technology suppliers like Intel and rival Advanced Micro Devices Inc. have reacted by designing chips that offer lower power consumption as well as greater performance. They argue that switching to new servers with such chips can save enough on power and labor costs to pay for upgrades in a few months.

Intel, for example, has overhauled its Xeon line of servers chips to include a model with the equivalent of eight electronic brains on one piece of silicon. The company estimates that a server with four such chips offers a 20-fold performance increase over an existing server with four single-processor chips; that means one new machine can take the place of 20.

Even before factoring in models based on Intel's newest Xeon chips, pricing for some server vendors is on the rise; the average price of Xeon-based servers sold by Hewlett-Packard Co., for example, rose nearly 12% to $3,993 from the second quarter of 2009 to the first quarter of 2010, market researcher Gartner estimates.

Customers have responded, in many cases paying up for servers with high-end chips that command higher prices. Mr. Erwin of ThePlanet says it moved swiftly this year to Intel's new technology, saving his company money on power and labor costs and providing greater performance to offer customers at a higher price.

Zach Nelson, chief executive officer of Web-based software provider NetSuite Inc., plans to use H-P servers with Intel's most-powerful chips in a new data center in Boston. "It maximizes our customer experience and reduces our cost," he says.

Other companies are adding different systems for different computing chores. Susan Shimamura, the vice president of operations at IAC/InterActiveCorp's Ask.com, says the company has traditionally bought only low-end Dell server systems for its Web search function. While continuing that practice, it recently decided to also buy higher-end machines for databases that analyze how people use Ask, she says.

Big-name server makers are not the only beneficiaries. To offer cloud-style services, Rackspace prefers little-known suppliers for attractively priced "white-label" servers "straight from the factory in Taiwan," says Lanham Napier, its chief executive.

Just how long the server-buying boom will last is unclear, amid economic jitters and the fact that cloud companies tend to buy servers in advance signing up customers.

"It's the build-it-and-they-will-come model," says Bryan Doerr, chief technology officer of Savvis.

But companies pursuing cloud computing say demand is so strong that they aren't worried about adding too much capacity. "This is a major tectonic movement," says Manuel D. Medina, chief executive of Terremark Worldwide Inc., which says its cloud business has been growing 30% sequentially each quarter. "There's zero chance of a bubble."

Intel's Larrabee Chip Dead?

PC Mag

Intel has canceled at least the first product iteration of its graphics chip, code-named "Larrabee," company representatives confirmed.

However, Intel remains committed to delivering multicore graphics products to its customers, a company spokesman said.

"Larrabee silicon and software development are behind where we had hoped to be at this point in the project," an Intel spokesman, Nick Knupffer, said in an email. "As a result, our first Larrabee product will not be launched as a standalone discrete graphics product, but rather be used as a software development platform for internal and external use."

Knupffer did not say why the Larrabee project was behind, or what, if any, milestones it had missed. In May, Intel claimed that it had "sharpened" Larrabee's release date to the first quarter of 2010, in response to a report that the company had delayed Larrabee's launch.

"While we are disappointed that the product is not yet where we expected, we remain committed to delivering world-class many-core graphics products to our customers," Knupffer added. "Additional plans for discrete graphics products will be discussed some time in 2010."

Intel originally launched Larrabee as a response to the general-purpose GPU products at Nvidia and at rival AMD, which were known as "Tesla" and "Fermi" at Nvidia, and "Fusion" at AMD. Although specific details were hazy, the first iteration of Larrabee was designed to counter the Fermi and Tesla products at Nvidia. Separately, Intel also developed a series of CPUs that began integrating graphics capabilities, the "Arrandale" and "Clarkdale" processors due next year. Over time, more of the Larrabee functionality was expected to make its way into succeeding generations of the "Dales" chips, as Intel sometimes calls them.

But AMD's own Fusion architecture won't debut until 2011, so Intel's postponement of its Larrabee project simply places the true integration of GPU and CPUs a bit farther out, while Arrandale and Clarkdale begin addressing the issue in 2010.

AMD's first "accelerated processing units" (APUS) won't debut until the "Llano" and "Ontario" chips are released in 2011. Nvidia, meanwhile, lacks X86 processor technology, although some have speculated that it might be pursuing a Transmeta-like, emulated instruction-set approach. Nvidia also recently clarified its position in regard to Larrabee's reported capabilities.

Knupffer's disclosure also casts some of Intel's recent portrayals in a new light. At the recent Intel Developer Forum, Intel demonstrated Larrabee by rendering a scene from a recent 3D game, Quake Wars. While the demonstration was graphically unimpressive, some saw it as merely an early iteration of test silicon. Others were more suspicious that what was shown represented something closer to the final product, which appears to have been more the case.

James Reinders, Intel's chief evangelist for its software products, also described Larrabee in what now might be seen in a more conservative light. In an interview published in October with ExtremeTech, Reinders was asked about Larrabee's role as a new, hybrid architecture. "Yes, it's radical, and it will take a number of years to get out there and for us to fully understand how to exploit it," he replied.

Reinders was also pressed for a performance comparison.

"Comparable? Absolutely, but when you introduce a new architecture you don't expect to be the top from day on," Reinders replied. "We haven't released any numbers, so I'm not sure where it will land or where the competition will be when we introduce it. But we believe in the architecture enormously because we think that this type of design will take us further than the current types of design."

Broadcom Buys Dune Networks

Hoping To Make 'Crazy Fast' Networking Chips

NY Times

Chip maker Broadcom said it has agreed to Dune Networks for $178 million in cash. The communications chip maker is pouncing on a startup that makes a scalable networking chip set for communications equipment that connects computers inside corporate data centers.

They may not be sexy. But these are the chips that make the Internet go faster, as they can transfer data between corporate computers known as servers at high speeds.

Dune Networks has developed a chip set that can handle data transfer speeds of 100 gigabits per second per port. It would be used in switching equipment for data centers with thousands of servers.

The deal is expected to close by the end of March. Investors in Dune include Alta Berkeley Venture Partners, Aurum-SBC Ventures, Evergreen Venture Partners, Jerusalem Ventures Partners, Pitango Venture Capital and US Ventures Partners. Dune was founded in 2000 and is based in Yakum, Israel, with U.S. offices in Sunnyvale, Calif.

Intel Cites Advance in Using Silicon in Data Products; Claim Is Challenged

Intel Corp. is claiming another advance in creating optical communications components from silicon, an effort designed to reduce the cost and increase the speed of transmitting computer data.

The company said it combined silicon -- the low-cost foundation for most computer chips -- with the element germanium to make a device called an avalanche photo detector that achieved record performance. Intel said the development marks the first time that a silicon-based optical component exceeded the performance of an equivalent device made from more costly conventional materials, such as indium phosphide.

But the importance of Intel's announcement was challenged by researchers at Luxtera, a closely held company that is already producing silicon-based optical components.

Optical communications involves encoding information on streams of light particles, generated by lasers. The technology brings big benefits in speed over standard electrical connections, and uses thin glass fibers rather than bulky cables. But optical connections are now mainly used for high-volume long-distance communications -- or connecting servers in massive supercomputers -- because key components often cost tens to hundreds of dollars each.

Researchers are hoping to drive those costs down to pennies by using materials found in conventional chips, a field known as silicon photonics. Intel, in particular, has been churning out a series of research papers describing prototype optical components made from silicon.

Its latest development, which was jointly funded by the U.S. Defense Advanced Research Projects Agency, is described in a paper in the journal Nature Photonics. Intel engineers collaborated with counterparts at Numonyx BV, a company comprised of former operations of Intel and STMicroelectronics NV. Experts at the University of Virginia and the University of California at Santa Barbara provided consulting and testing, Intel said.

Photo detectors are used to sense and amplify light pulses generated by lasers. The new prototype detector achieved a "gain-bandwidth product" of 340 gigahertz, which is the highest result recorded to date on that key metric of detector performance, said Mario Paniccia, who directs Intel's photonics-technology lab and holds the title of fellow.

Improvements in detector performance could be exploited in different ways, including boosting the speed data is sent, increasing the distance a signal goes or reducing the energy needed to send a signal a constant length, Mr. Paniccia said. "And we believe we can continue to improve the performance," he said.

Intel initially expects silicon-based optical components to send data between servers in a computer room and between chips in a system, though it later hopes to have optical connections inside its microprocessor chips, too. Mr. Paniccia said silicon-based detectors also could find uses outside communications, in applications such as optical sensors, cryptography and medicine. He didn't give a precise timetable for turning the new components into products, but indicated it would take several years to perfect the technology.

Meanwhile, Luxtera is "ramping up" production of its silicon-based optical components, said Greg Young, chief executive of the Carlsbad, Calif., company. He said the performance described in Intel's paper is "tremendous." But Mr. Young contends that Luxtera researchers actually were the first to top the performance of indium-phosphide photo detectors in research results published more than a year ago.

Mr. Young also asserted that Intel's technology is incompatible with conventional semiconductor-production processes, so it couldn't be used in a so-called wave-guide detector to work alongside other components on one piece of silicon.

Mr. Paniccia said Intel is developing a wave-guide version that could be integrated on a chip.

And many applications, including inexpensive fiber-optic links to homes, don't require wave guides, said John Bowers, a silicon photonics expert and professor of electrical and computer engineering at University of California-Santa Barbara. "It's a huge paradigm shift," he said of Intel's results.