Intel’s upcoming 10nm “Ice Lake” processor won’t deliver any more cores and threads than its current 8th-gen parts, and at slower turbo clock speeds. But what Intel is calling its 10th-gen Core chip here at Computex offers numerous improvements across the board: performance boosts across CPU, graphics, and AI tasks, plus platform-level enhancements like “Wi-Fi 6 Gig+” that offers more bandwidth than your wired router.
The Ice Lake-based 10th-gen processors will be Intel’s first widely available 10nm Core chips and the company’s first major chip redesign since 2015’s Skylake architecture. They're now shipping, Intel's Gregory Bryant, senior vice president and general manager of the Client Computing Group, is expected to say in his Computex address.
Since January, Intel has said that the first Ice Lake chips would ship to customers this quarter, specifically in June, as part of an initial crop of 30 notebooks which will appear on store shelves by this holiday season. Ice Lake will replace the current “Whiskey Lake” U-series and “Amber Lake” Y-series Core chips with 9W, 15W, and even some 28W versions.
While Intel’s holding back specifics on individual chips, we still now know quite a bit about the Ice Lake family as a whole: Core i3, i5, and i7 chips will be offered, with up to 4 cores and 8 threads—the same as Whiskey Lake. Ice Lake’s turbo boost speed of up to 4.1GHz is far less than the Whiskey Lake-based Core i7-8565U’s 4.6 GHz boost, but a new, improved method of turbo management may mitigate that. Finally, the return of (now Gen11) Iris Plus integrated GPUs could be a big deal that widens the options available for notebook gaming.
Performance numbers, though, are still vague. The most concrete estimate of Ice Lake’s performance still comes from Intel’s investor meeting: two times the encode and graphics performance of Whiskey Lake. But there’s a wild card. Remember how Intel’s 7th-gen “Kaby Lake” processor was optimized for video playback, because of the rise of YouTube and Netflix? Ice Lake makes similar strides in AI. You’ll also see some unexpected perks in how Intel implements an enhanced Wi-Fi 6 Gig+ wireless technology and Thunderbolt 3, too.
In his Computex keynote, Bryant also announced 9th-gen vPro chips, with up to 8 cores and 16 threads reaching up to 5GHz on desktop and up to 4.8GHz on mobile -- even a Core i9. Intel also launched 14 new Intel Xeon E processors for mobile and desktop workstations. Bryant also disclosed that a new series of "X-series" Ice Lake chips will debut this fall, alongside increased memory speed and an updated Intel Turbo Boost Max Technology 3.0. Bryant also showed off the new Intel Performance Maximizer, an automated overclocking tool.
How fast is Ice Lake? Here’s what we know
Much of Ice Lake’s design was informed by Project Athena, Intel’s redesign of the ultrabook. Chris Walker, vice president and general manager of the Client Computing Group at Intel, said that the modern workforce has to be used to switching back and forth between a power plug and a battery, between work and life. Consumers value connectivity and long battery life, Intel says, but they turn to the PC to focus on a task, wherever they may be.
“It’s not a separate work PC and a separate home PC, it’s one—and we’ll start to see that coming into our partners’ design language,” Walker said.
In some sense, that lessens the impact of Ice Lake as a CPU, and emphasizes the broader appeal of Ice Lake as a platform. The fact remains, however, that so much of the PC still revolves around the processor, and about Intel’s struggles to move beyond 14nm.
There’s an old adage in the chip industry: never launch a new processor design on a new manufacturing process, because of the increased risk of costly manufacturing errors. Intel seems to believe it’s dodged that risk, as last year it launched the untested 10nm process via Cannon Lake, a relatively unknown Core i3 chip that Intel shipped to Asia last year as part of a Lenovo notebook. Becky Loop, an Intel fellow and the Chief Client Architect, said that every piece of IP on Ice Lake was “touched” and optimized for 10nm.
Ice Lake, however, is built upon the new ”Sunny Cove” (now apparently spelled Sunnycove) core design that Intel disclosed late last year. Sunnycove was designed to be deeper, wider, and smarter, in the words of one executive. In reality, that means that Sunnycove chips like Ice Lake can perform more operations in parallel (5, versus 4 in Skylake) and with an increase of 8 to 10 execution ports. Sunnycove’s level 1 data cache has increased from 32KB to 48KB, and the level 2 cache from 256KB to 512KB. Caches are used to store frequently used instructions, versus having to hunt them down from system memory. Think about them in the same way that large, deep pockets are useful places to store things you frequently need.
In general, Intel’s claiming that the Sunny Cove-based chips will perform, on average, 18 percent more instructions per clock cycle than the Skylake architecture across a variety of benchmarks. Intel’s also added instructions for specific tasks, such as crypto performance, vector AES, and security enhancements such as User Mode Instruction Prevention.
Meet Ice Lake’s Gaussian Neural Accelerator
Intel, though, made at least two specific improvements for AI: adding DL Boost (Deep Learning Boost) instructions and a Gaussian Neural Accelerator, a dedicated logic core designed to speed up AI functions. Don’t think of AI in terms of digital assistants like Cortana or Alexa. Instead, both of Ice Lake’s new technologies are designed to speed up tasks such as background blurring in Microsoft Teams, semantic search in Microsoft Photos, or recognition and transcription of a business meeting.
It’s the latter task that Ronak Singhal, an Intel fellow and director of the CPU Computing Architecture, called out the Gaussian Neural Accelerator for. It runs constantly in the background, but at very low power, performing a computationally intensive task without blazing through your battery. Sunny Cove, Intel claimed, offers between 2 to 2.5 times more performance in AI compared to the “Whiskey Lake” Core i7-8565U.
Executives also said that Ice Lake uses AI to manage boost performance, what it calls Intel Dynamic Tuning 2.0 with Machine Learning. Unlike older generations, which used more of an “all or nothing” approach to apply the overclocking capability in a chip’s Turbo Boost, Intel’s Dynamic Tuning feathers the accelerator a bit, boosting sharply, then voluntarily backing off, while trying to determine the “just enough” level of overclocking performance and prolong that as long as it can. Intel hopes that this smart performance management will overcome the lower turbo boost threshold, though it remains to be seen if that’s true.
Unfortunately, Intel said very little about Ice Lake’s actual power consumption, though the shrink to a finer 10nm process should theoretically lower it, extending a notebook PC’s battery life. Ice Lake also includes fully integrated voltage regulators, Loop said.
Next page: Graphics and Wi-Fi improvements, Ice Lake vs. Ryzen 3000