Intel took the wraps off its newest processors for data centers Tuesday, promising a significant boost in performance compared to its previous generation of chips thanks to a new design that improves how data flows across the chips.
The Intel Xeon Skylake processors have been out for a while with some of Intel’s more friendly customers, but they are generally available as of Tuesday. Navin Shenoy, the new manager of Intel’s data center group, called the launch the company’s “biggest data center platform announcement in a decade” at an event in New York.
Intel dominates the market for processors used by the big cloud computing companies as well as companies that manage their own infrastructure, with more than 95 percent of this business. Google has been running these processors on production workloads for customers since February, it said during the event, and Amazon Web Services also touted its support for the new chips when it comes to machine learning workloads.
For this generation of its flagship Xeon chips, Intel changed the way it manages the flow of data across different parts of the processor. The chip maker used to route data around its multicore processors with a “ring architecture,” which forced data to basically do a lap around the chip to reach a processing core or memory interconnect on the other side, said Lisa Spelman, vice president of Intel’s Data Center Group. Now it is employing a “mesh architecture” that lets data flow directly between cores.
That change, coupled with new instructions for accelerating workloads around artificial intelligence and security, resulted in a 1.6x performance gain compared to Intel’s last generation of chips, Shenoy said. “That’s the largest generation-on-generation performance improvement in the past decade,” he said.
While Intel still enjoys the lion’s share of this market, rival chip makers are getting ready to challenge its dominance, aided and abetted by some of the same big cloud providers that endorsed the Skylake chips Tuesday. Google has worked with IBM on Power chips for data centers, Microsoft (conspicuously absent from Intel’s launch event) and Baidu announced plans to deploy a new server processor from Advanced Micro Devices last month, and other chip makers are still exploring the idea of bringing ARM-based processors to this market. (Microsoft reached out Wednesday to note they do plan to support the new processors, even if they didn’t record a marketing video for the launch event.)
But unless Intel falters on the manufacturing side, this new generation of processors will wind up powering the vast majority of data center servers for the foreseeable future.