In the last week, we’ve had several dramatic turn-arounds in the CPU space. First there were leaks about the nineteen models AMD was going to offer with Ryzen, then there were pricing and supposed benchmarks for Ryzen processors. Then, out of nowhere, a rumour popped up that Intel was already sampling Kaby Lake refresh chips with better performance and thermals, along with a strange “Xeon Gold” leak that suggests they’ll start selling Xeon E5 and E7 chips from the Broadwell-E family for the X99 platform. In amongst this crazy week was a rumour that Coffee Lake, their next evolution of the desktop processor, might be launching early as a response to AMD’s Ryzen processors, and yesterday Intel confirmed that this was true – Coffee Lake is now scheduled for a 2H 2017 launch, rather than a 2018 launch. But wait, there’s more!
Coffee Lake, according to old roadmaps, was supposed to be the optimisation process of Intel’s new product cadence strategy. Instead of tick-tock, as has been the norm for over a decade, the new pattern is taking an old design and mating it to a new process (the tick), then launching a new design on that same process, now matured (the tock), and then an optimisation of the previous release with higher clock speeds. Up to now, this has been the case with the move from Broadwell, to Skylake to Kaby Lake. In a technical sense, Coffee Lake is an optimisation of an optimisation, which makes no sense, but it is what it is.
In its keynote address at their Investor Day event recently, Intel confirmed that Coffee Lake now is bumped up to a 2017 release, and promises a 15% performance boost over Kaby Lake. Although they didn’t state it, it makes sense that Coffee Lake is also on the same socket, seeing as the Intel 200-series chipset just launched this past month.
On top of this, Cannon Lake, the mobile successor to Intel’s Skylake U and Y processors, isn’t going to be fabricated on a 10nm node, which means that Intel’s plans to launch 10nm products in 2018 are also off the cards. This has been quite the week for them, hasn’t it?
In addition, future uses of new or existing high-end process nodes according to Intel will be “fluid” in their use, which means that Intel will not be trying to move to a new node on every product from the start. 10nm is now hitting their datacenter products first, where the power and cost savings are more easily realised by Intel’s partners and customers.