One thing that has been frustrating about chip giant Intel's (NASDAQ:INTC) current seventh-generation Core processors for personal computers -- and what will be frustrating about its upcoming eighth-generation Core processors -- is their apparent lack of architectural innovation.
A brief review may be in order. Back in the second half of 2015, Intel introduced a processor architecture that it called Skylake, which brought a lot of new technologies to Intel's chips: redesigned processor and graphics cores, improved multimedia cores, new (to PC chips, anyway) technologies like integrated sensor hubs and image signal processors, a new display engine, and other goodies.
In late 2016, Intel rolled out a successor to Skylake known as Kaby Lake. which largely reused the same cores and technologies as its predecessor -- though it did come with an improved media engine. However, thanks to an upgraded manufacturing technology, Kaby Lake chips were more efficient and more powerful than their Skylake counterparts.
Later this year and in the first half of 2018, Intel will reuse that same basic Skylake architecture in its eighth-generation Core processors (to be called Kaby Lake Refresh and Coffee Lake, depending on the product segment). Once again, Intel plans to bring enhanced manufacturing technology to bear, and this time around, it will boost the number of processor cores included in its chips.
The good news is that Intel should be able to deliver on its promise of newer, better products each year. The bad news is that, for a third product generation in a row, Intel isn't making fundamental improvements to its key processor and graphics technologies.
At Intel's Technology and Manufacturing (TMG) Day, one analyst asked the following question about this very topic:
Why not introduce a new architecture design sooner than Ice Lake, which is 2019? You guys used to do architecture redesign every two years and now it's, between Skylake and Ice Lake, it's going to be like four years unless I'm mistaken. Architectural design used to be independent of nodes. So why not introduce one sooner than later?
Let's look at what Intel executive Murthy Renduchintala had to say.
Deconstructing the answer
Renduchintala began by arguing that in this discussion "we need to be really precise about the expansiveness of terms like architecture."
He then went on to clarify that "we can talk about new ways of interconnecting the various [intellectual property] blocks within a [system-on-a-chip] to get better performance, but let's also be clear that what we're also doing is upgrading key pieces of [intellectual property] such as CPU or GPU, looking at different configurations of clocks and memory bus access speeds."
This is where Renduchintala lost me.
The problem isn't that Intel isn't changing up how it hooks up its CPU cores, graphics cores, and other pieces of technologies; it's precisely that Intel hasn't really changed those intellectual properties over the course of several generations that's the problem.
Renduchintala then went on to say that "there isn't really a panacea where you basically move toward a complete new architecture that in and of itself is necessarily going to give you all of those gains."
I agree with him when he says that to get the kinds of product-level performance increases that the company aims to deliver on an annual basis, Intel's teams need to deliver on a "mixture of modification at the chip level construct, the improvement of [intellectual property], looking at the transistor process evolution, and creating a compound chemistry of that actually give[s] that performance improvement."
Where Renduchintala and I disagree is on whether Intel has really been fully exploiting the product-enhancement opportunities available to it.
What I would like to see Intel do
Intel has done a good job of reshaping its chip development methodology to allow it to deliver new products at an annual cadence, even if it can't move to tinier manufacturing technologies every two years like it had hoped to do.
However, if Intel could deliver annual improvements in the underlying chip architectures themselves -- that is, if it could improve how each individual intellectual property block (like CPU and graphics) processes data -- I think that could help it deliver even larger generation-over-generation product improvements.
What I'd like to know is why Intel left its key intellectual properties largely unchanged between its sixth-generation and eighth-generation cores. Was it simply due to time-to-market concerns, potentially owing to the seventh- and eighth-generation Core products being added to the company's product roadmap relatively late in the game?
Or would Intel have stuck with those same intellectual properties even if it had known many years in advance that it would need to stay on derivatives of its 14-nanometer technology for a while?
Perhaps Intel will shed some light on these topics at some point in the future.