Smooth Stone Vows To Nearly Eliminate Server Power Use In Data Centers

August 16, 2010

Secretive chipmaker Smooth Stone came out of stealth today with an announcement of $48 million in funding and a vow to virtually eliminate server power use in data centers.

The sheer size of the Austin company’s first venture round provides a useful validation of its business plan. The funding drew investors Battery Ventures, Highland Capital Partners, Flybridge Capital Partners, Texas Instruments, the Advanced Technology Investment Company and ARM, which had already provided money to the company and has a seat on its board. (Smooth Stone had previously raised $3 million of seed capital.)

The company is among a new wave of start-ups aiming to cut data center power use – specifically server energy consumption.

CEO Barry Evans declined to provide details about its development and delivery of an ARM-based chip. But he said the goal is to deliver a “disruptive” technology that will eliminate “almost all” of the power demands of a data-center server. ARM chips are generally used in low-power devices, such as cell phones and handhelds.

“We want to take power out of the equation” for data center managers, he said in an interview. “The problem is becoming so acute” – along with data-center rack space.

Other companies addressing the market include SeaMicro, which promises one-quarter the power and one-quarter the space of a typical Intel-based Xeon server. The company’s SM1000 server uses 512 Atom chips, another Intel chip made for netbooks and other low-power devices.

A third company apparently making high-efficiency processors for servers is Angilux, which Google bought earlier this month.

Smooth Stone’s business model appears to be focused on chip making, leaving the development of servers to partners and customers. The company isn’t offering much detail on its plans other than to say closing the $48 million round required it to have all the important pieces lined up, from technology to market validation.

Smooth Stone has been engaged with the market for some time, says Evans.

He adds that the money will allow it to complete its chip development and double the size of its development team over two to three months. That could hasten its move into the market.


SeaMicro’s Minor Revolution In The Data Center

June 14, 2010

Not since the advent of virtualization has the data center faced such an opportunity for change.

Low-cost, ultra-low-power servers – sometimes called microservers – may finally get a jolt of legitimacy.

On Monday, secretive Santa Clara start-up SeaMicro formally launched its long-awaited remake of the x86 server: a 512-processor box that doesn’t use Intel’s ubiquitous Xeon chips but low-power Atoms instead. Atoms are the processors sold in $300 netbooks – giving rise to the observation that SeaMicro’s SM10000 is really just a collection of netbooks stuffed in a one box.

Expect vendors such as Dell to begin making microservers. SeaMicro claims it will cut power use by 75 percent.

It’s an observation that is essentially true. The result is a server that uses one-quarter the power and takes up one-quarter the space while performing the same amount of work. CEO Andrew Feldman says Atom is three times more efficient in performance per watt than Xeon. The reason is it can better power down when not in use and doesn’t waste energy trying to anticipate future workloads, as Xeon does.

That is why the SeaMicro box is better suited to the Internet, where traffic is bursty and generally only places lightweight demands on a server.

The SM10000 is the brainchild of Gary Lauterback, a former AMD fellow and Sun Microsystems engineer, and “is an enormous transformation of the data center,” claims Feldman.

He may not be exaggerating. Zeus Kerravala, a long-time tracker of the server industry at Yankee Group, says: “As an analyst I am often skeptical of technologies people tout as revolutionary, but this one I was really impress with.” If Dell and other top tier vendors aren’t already thinking about microservers “I’d be surprised,” he says.

In truth, SeaMicro isn’t the first company to conceive of low-power servers or ones running Atom. Super Micro Computer launched a rack-mounted Atom blade last year, and Hewlett-Packard markets a $400 MediaSmart home-server with Atom. In Austin, TX, Smooth Stone is working on technology to bring even lower-powered ARM processors, those in many cell phones, to the server market.

Improved efficiency is what motivated cloud-computing vendor Rackable Systems to make use of small servers with modest power to handle fluctuating Internet workloads.

However, SeaMicro hopes to take Atom boxes a step further. The company built into the SM10000 a 1.28-terabit communications fabric powerful enough for a super computer and installed a custom ASICs to handle the complex load balancing for 512 processors. A single box can replace 40 dual-socket, quad-core servers, two Ethernet switches and two terminal servers, says Feldman.

It also shrank the size of processor motherboards to the size of a credit card, taking off unnecessary components and reducing the power draw.

According to IDC, the package may catch on with Web 2.0 companies. Companies spend $27 billion globally a year buying energy to run their servers, the research firm says. Most would die to reduce the bill.

“I think it is a radical approach” that Web 2.0 companies will quickly adopt for their public clouds, says research analyst Katherine Broderick.

The SeaMicro, which raised $25 million from backers including Khosla Ventures, Draper Fisher Jurvetson and Crosslink Capital and received a $9.3 million Department of Energy grant, will make the box available in July. Selling for $139,000, it is likely to begin earning its investors a return.


Data Center Energy Pressures Could Ease Next Year

September 22, 2009

Data centers are a necessity of the modern 21st Century economy. They keep companies from Google to Procter & Gamble in business.

Starting next year, servers designed to sit at the center of these giant info-warehouses will begin to show some dramatic reductions in energy use.

The result could be big gains in the energy efficiency of data centers. Of course this means replacing computers already in use with new models, something most firms will not do. But overtime as these energy efficient units find their into use spiraling energy costs could for the first time decline.

A Samsung laboratory experiment found that servers expected on the market next year will burn 38 percent less energy than those available today. Part of the reason is the adoption of DDR3 memory chips made with a new generation 40-nanometer manufacturing equipment. The smaller circuitry requires less power. (Samsung is a manufacturer of the chips.)

But the machines also are expected to adopt Energy Star guidelines released by the EPA in May. The guidelines suggest that 30 percent energy savings are possible.

Sylvie Kadivar, an associate director of strategic marketing at Samsung, said Tuesday that a reduction in server energy use is expected. “We obviously believe it will be dramatic,” she said,

The poser savings from a 2GB 40-nanometer DDR3 memory chip could by itself be 73 percent, she said.

How much energy will be saved worldwide from this new generation of server? It is hard to calculate since adoption rates and market growth are hard to predict. But the development does suggest a change in the trend line is possible.


Cisco Hopes To Commoditize HP, Sun And IBM

March 16, 2009

Sun Microsystem’s famous proclamation that “the network is the computer” may be more true than ever.

And Cisco System may be the firm to prove it. Cisco announced on Monday its first foray into servers – unveiling an Intel-based blade server designed to go inside a Unified Computing System that includes storage, networking and virtualization.

Cisco wants to deliver the whole enchilada with its Unified Computing System, says Miko Matsumura

Cisco wants to deliver the whole enchilada with its Unified Computing System, says Miko Matsumura

So why buy a server from Hewlett-Packard, Sun or IBM if you can get it fully integrated from Cisco? If the server offers better performance, that might be a reason.

But if instead hardware becomes a commodity and the real “special sauce” is in how a company delivers data and services to its users, then it might not matter.

Cisco has already been expanding its reach for several years. It pushed into home networking, consumer goods, set-top boxes, IP phones. Servers is another logical step.

“They are really trying to deliver the whole enchilada,” says Miko Matsumura, deputy CTO at software maker Software AG.

Delivering the complete meal comes with several benefits. First, customers have only one company to work with and blame if something goes wrong. Integration is Cisco’s problem, not the buyer’s.

Second, as Web applications become more complex and software mashups more common, computing systems are having an increasingly difficult time communicating with one another. Communications are hampered as subtle variations appear in the typical seven layers of protocols networks use to communicate.

Cisco has the potential to solve this by integrating networking gear, servers and storage with a layer of virtualization, says Matsumura. “The router is a virtual router. The storage is virtual storage.”

Users (and IT workers) see less and less of the complexity.

So why buy a stand-alone server from H-P?


Multi-Core Chips Multiplying Out Of Control

January 28, 2009

Multi-core chips for several years have delivered more performance than most software today can use.

Many multi-core chips outpace the capabilities of software

Many multi-core chips outpace the capabilities of software

To truly harness the power of chips that have more than two computing cores, software needs to break its tasks up into pieces that can be acted on in parallel, each by a different core.

It’s a performance gap Gartner sees increasing in the years to come. This from Gartner:

On average, chipmakers get double the number of processors on each new chip approximately every two years. With this doubling of processors and potentially more threads per core, a server with the same number of sockets for chips gets twice as many processors. In this way a 32-socket, high-end server with eight-core chips would deliver 256 processors in 2009. Two years later, with 16 processors per socket, the machine swells to 512 processors. Four years from now, with 32 processors per socket, that machine would host 1,024 processors.

“Most virtualization software today cannot use all 64 processors, much less the 1,024 of the high-end box,” said Carl Claunch, a distinguished analyst at Gartner. “There is a real risk that organizations will not be able to use all the processors.”


Follow

Get every new post delivered to your Inbox.

Join 32 other followers