Intel Integrating Network Fabric into Xeon

Intel plans to integrate a network fabric controller into its Xeon microprocessors within the next few years, reports revealed Monday.

Raj Hazra, vice president of the company’s Intel Architecture Group, revealed Intel’s general plans—but precious few details—to publications such as The Register.

Intel will kick off its Intel Developer Forum this week in San Francisco, where Diane Bryant, where general manager of the Intel Datacenter and Connected System Group, will speak talk about Intel’s data center plans later on Sept. 11. But the network fabric, the critical I/O of a data center, has become increasingly important to the chip giant.

“There is a difference between networks and fabrics, and while there is a place for networks, they lack certain optimizations that fabrics have,” Hazra told the Register. “Some applications need purpose-built interconnects, and fabrics look at compute and storage nodes as partitioned logical resources rather than as separate units of compute and storage.”

Problems are becoming superscalar across multiple machines, he added, and that is driving new approaches of adding bandwidth and reducing latencies in that bandwidth: “The fabric interconnect has become what was the system bus or processor interface.”

If one looks strictly at the race between AMD and Intel, then it’s clear that network fabrics may evolve into the next big arms race between the two. At a Sept. 10 event, AMD chief technology officer Mark Papermaster called the Freedom Supercomputer Fabric—which his company obtained with the acquisition of SeaMicro—one of the most important pieces of the acquisition, as well as a key technology for AMD over succeeding generations.

That fabric connects hundreds of the SeaMicro min-motherboards together in a torus configuration at 1.28 terabits per second. And dense, energy-efficient computing is the name of the game for SeaMicro, whose I/O virtualization technology can effectively cut out the need for everything but the CPU, memory, and the SeaMicro ASIC.

AMD hasn’t announced a roadmap for the Freedom Fabric, but it plans to develop future iterations of the technology over successive generations, Papermaster added. (AMD’s SeaMicro announced its SM15000 server late on Sept. 10, which brings the Freedom Fabric outside the box to storage.)

Intel hasn’t said whether it plans to develop a closely-coupled network fabric chip, or more tightly integrate it inside the CPU module, or, eventually, on die. But earlier this year, the IT giant spent $140 million to buy the “Gemini” interconnect from Cray, as well as the rights to the engineers and the intellectual property backing them. That’s a good indicator that Intel now has the technical chops to back up its promise.

 

Image: kentoh/Shutterstock.com

Post a Comment

Your email address will not be published.