More on the same topic
---------------
Intel, long known for cookie-cutter computer chips, is stepping up efforts to let companies customize its processors.
The Silicon Valley giant is using an industry conference in San Francisco Wednesday to announce an unusual plan to package its standard server processors alongside a different class of chip, which can be electrically programmed to handle specific calculating functions. Intel says the combination can let Web companies and other customers achieve dramatic performance gains on some tasks.
Intel is offering other customizing options for its widely used Xeon line, including building in specialized circuitry that customers design themselves. But that process could take many months.
The company says the new two-chip combination--which includes what the industry calls an FPGA, for field programmable gate array--allows customers to quickly adapt the functions of those chips in a speedy process akin to changing a piece of software. They can also modify those programming instructions after the servers are installed with no hardware modifications, if their software or business problems change.
"The pace of software innovation is crazy fast," said Diane Bryant, a senior vice president who is general manager of Intel's data center group. "This is a great way for the silicon to keep up with software innovation."
FPGAs have been sold for years by chip makers such as Xilinx Inc. and Altera Corp. Ms. Bryant declined to disclose Intel's partner on the latest effort, or when the resulting products would hit the market.
Some customers already plug FPGAs into server systems to accelerate certain workloads. That can bring a tenfold speedup in some jobs, Bryant says.
But placing the FPGA and a server's central processor together in the same package can be twice as fast, she says, because they exchange data on a faster kind of connection and share memory more efficiently.
Bryant says some customers she won't identify are trying out the dual-chip offering. Last year, Intel disclosed that it has modified Xeon models to benefit customers that include Facebook and eBay, with a total of 15 customized products developed. This year, Bryant vows to deliver more.
Meanwhile, Intel on Wednesday is also disclosing the official name--Xeon D--for a new multi-function version of the product line expected out next year. Where Intel has such "systems on a chip" for low-power applications already, based on its Atom family, the Xeon D will be beefier "SoC" and will use the company's latest production process.
Why all the effort? After all, Intel commands roughly 97% of industry shipments of server chips.
For one thing, big Web companies including Google and Facebook have made noises about experimenting with new chip technologies, including the Power line that originated at IBM and the ARM Holdings designs that many companies sell for mobile devices.
For another, Bryant is pushing to get standard processors from Intel into networking, data storage and other applications that now typically use chips based on Power or MIPS designs (MIPS Technologies, once a high-profile Silicon Valley company, fell by the wayside in the mainstream computer market and was purchased by the British company Imagination Technologies Group last year). Backers of ARM are racing to grab those same applications.
"As we have seen many times during our history, times of disruptions call for a choice for organization to embrace innovation or risk losing their position in the market," Bryant noted in a blog post drafted to accompany her announcement, which came during an appearance at the Gigaom Structure conference.
