The Next Evolution of Servers

Nine big names in computing are establishing an effort to shape the evolution of servers, introducing a challenge to Intel Corp.

The firms said Friday they would support technology developed by International Business Machines Corp. that promises to transfer information among parts in servers up to 10 times faster than present circuitry. The firms hope to persuade others to adopt the technology.

The new technology plans to conquer an emerging bottleneck in managing some computing jobs. While conventional microprocessors are becoming faster, their computations are frequently delayed by the need to get information from nearby memory chips, image chips or other parts used to manage specialized jobs.

  • The problems are particularly acute, business executives say, as computers tackle vast sets of information like images or speech patterns.
  • “We are going to need to bring new computer designs, new technologies to carry to service the needs of these new modern workloads,” said Brad McCredie, an IBM vice president.

As well as IBM, the other participants in the new attempt are Alphabet Inc. ‘s Google, Hewlett Packard Enterprise Co. , Dell Technologies Inc., Advanced Micro Devices Inc., Micron Technology Inc., Nvidia Corp. , Mellanox Technologies Ltd. and Xilinx Inc. Some firms will also be pushing for more technical diversity.

IBM has particular reasons to push for alternatives.

After years of reserving Power-based chips for its own servers, IBM in 2013 began licensing the technology to other manufacturers of chips and systems.

Google, which relies heavily on x86 systems it designs itself, became a leading member of that group. In April, the web search giant announced a strategy to develop a server design based on IBM’s latest Electricity chip. A Google spokeswoman said it also intends to use the IBM-developed communications technology as part of that design.

That technology is called CAPI, for coherent accelerator chip interface, and the new group boosting it is dubbed OpenCAPI. Coherent refers chiefly to technology that enables multiple parts—such as cards comprising graphics chips and microprocessors—to share the same pool of memory chips.

Analyzers say that approach has advantages in many computing jobs compared with the typical connections now used in most servers, which need graphics cards to use separate memory chips. But they question whether OpenCAPI will develop Intel’s support.

“This isn’t something that Intel would need to participate in,” said Patrick Moorhead, founder of the market research business Moor Insights & Strategy.

An Intel spokesman said its own technology “provides our customers with the finest combination of alternative, performance, and total cost of ownership.”

IBM said servers and associated products predicated on OpenCAPI are anticipated in the second half of 2017.