Intel executives spent hours on Wednesday publicly explaining their plans to expand chip sales for servers and cloud data centers amid a resurgent threat from Advanced Micro Devices. Unable to offer the simplest answer, however, Intel’s strategy session failed to entirely reassure investors, analysts, and customers.
The challenge for Intel, as it has been for several years, is that it can’t rely on easily improving chip performance and efficiency like it did previously—by printing smaller versions of the billions of tiny circuits in each chip. At least for Intel, decades of steady progress predicted by Moore’s Law—the maxim that chips can hold twice the number of transistors every few years— has stalled.
Intel’s plan to shift from manufacturing 14 nanometer circuits to 10 nanometers — the industry’s not-precisely-accurate shorthand for the scale of fitting transistors on a chip—first targeted for 2015 won’t happen until late next year for regular PCs and in 2020 for servers. That’s a major delay compared with AMD (AMD), which is sprinting ahead.
Earlier this year, AMD shifted from producing 14 nanometer to 12 nanometer circuits. And it has already shown partners a 7-nanometer version of its Epyc server chip, code-named Rome, which will go on sale next year.
Without mentioning Intel by name, AMD CEO Lisa Su alluded to her rival’s problems while speaking to analysts last month. “We see incredible opportunities ahead based on the competitive positioning and customer interest in our upcoming 7-nanometer products,” Su said.
Intel’s event on Wednesday, in Santa Clara, Calif., was intended to show that its “competitive positioning” was better than outsiders realized. Even without the jump to 10 nanometer sizes, Intel’s chips would offer compelling performance improvements, the company argued.
Intel’s next generation server chips, called Cascade Lake, coming later this year, and the next upgrade, called Cooper Lake, coming in 2019 will both be at 14 nanometers. They will include Intel’s usual “step function” of faster processing times from improved designs and new features, the company said without specifying how much faster. The Cooper Lake chips will also be compatible with the same server designs as the 10-nanometer Ice Lake line finally coming in mid-2020, making it easier and cheaper for customers to upgrade to the faster chips.
Intel’s stock, which plunged 9% on July 27 after details of the latest 10-nanometer chip delays emerged, gained only about 2% since Wednesday’s event. For the year, Intel’s shares are up 12%, versus a gain of 88% for AMD and 33% for fellow rival Nvidia.
Performance leadership
The presentations on Wednesday aimed to spell out some of the new features and improvements in the 14-nanometer chips.
“We’ve been talking a lot about 2019 and having sustained, continued data center performance leadership,” Lisa Spelman, Intel vice president and general manager of its Xeon server chip line that includes the new, upcoming chips, tells Coins2Day. “We are building to that and delivering that across the entirety of the road map that we showed.”
Spelman wouldn’t provide more detail about the improved performance of Intel’s upcoming chips. That will come at a later formal introduction, she promised. At last year’s introduction of the Xeon Scalable line, Intel customers said they’d seen a 30% improvement, for example.
This year’s Cascade Lake line will also include hardware changes to improve security to address the Spectre and Meltdown vulnerabilities revealed earlier this year that tarnished the entire chip industry’s reputation, she said.
Upcoming Intel chips will also have special features to speed up artificial intelligence and machine learning tasks, a growing need in data centers that rival Nvidia has rushed to meet. Intel (INTC) provided no new details about its own early effort to create graphics chips, an area which it ignored for years, to compete with Nvidia (NVDA) more directly.
Analysts unimpressed
Analysts were less than impressed by Intel’s failure to address the performance of its chips versus the competition . “We would have loved to have gotten more direct insight into the 14 nm strategy for holding share in server microprocessors,” Morgan Stanley analyst Joe Moore noted on Thursday. Not helping matters is that Intel is currently looking for a permanent CEO since Brian Krzanich resigned in June. “We would expect the resolution of the CEO search to provide some clarity,” Moore added.
The presentations “did not instill a lot of confidence on how they will maintain performance leadership across the entire product line given delay of 10 nm and no public 7 nm timing (likely 2022),” Barclays analyst Blayne Curtis concluded.
At this point, Intel “has mostly lost control” of the debate given its repeated delays, Bernstein analyst Stacy Rasgon wrote after attending the event. “The evolution of the competitive environment around the CPUs themselves remains an open question,” he wrote on Thursday. “The sessions did little to improve the company’s narrative around…forthcoming relative performance characteristics.”
Analyst Tim Arcuri was more excited about some of the server products Intel showed beyond just CPUs, like memory chips and storage systems which could lift sales and offer features competitors could not. “Although there was really nothing new, we were encouraged that INTC is ‘surrounding’ the data with a myriad of data-centric products,” Arcuri wrote on Thursday. “Ergo, there is more headroom here for growth than most investors appreciate.”
