In an article published earlier this month, I speculated that one key performance enhancement that Intel (INTC 0.20%) could bring to its upcoming Kaby Lake processor family over the just-released Skylake processor family is a boost in clock frequency.
In this case, it would make sense that even though Intel wouldn't be able to get more performance-per-clock (since it would be the same Skylake CPU core in Kaby Lake), it could get more performance by increasing clock speeds.
Although I still believe that this is the route that Intel is likely to take, I do think there's a possibility that Intel could do something else to improve CPU performance in Kaby Lake over Skylake -- use the Cannonlake CPU core.
Intel did it with the GPU, why not the CPU core as well?
It is widely believed at this point that Kaby Lake uses the graphics engine that Intel is expected to use in its 10-nanometer Cannonlake processor, but ported to Intel's less advanced 14-nanometer manufacturing technology.
Although I believe that a CPU core like Skylake/Cannonlake is more difficult to port across manufacturing processes than a graphics engine, it may not be so farfetched to think that Intel might choose to invest in implementing the Cannonlake CPU core in Intel's 14-nanometer manufacturing technology.
After all, when we're talking about a product family that will literally generate tens of billions in revenue over its lifetime, the costs to do such a "backporting" of Cannonlake to 14-nanometer probably aren't so high.
Indeed, by doing this, Intel would be able to deliver a reasonable per-clock performance improvement (I'd guess around 5% given the historical performance-per-clock increases Intel has delivered with "ticks" in its "tick-tock" cadence) and potentially even greater clocks than what Intel delivered with Skylake by virtue of the 14-nanometer manufacturing process becoming more mature over the next year or so.
Would such a move be worth it?
During Intel's most recent earnings call, CFO Stacy Smith made a simple, yet powerful, statement regarding what drives gross profit margins at semiconductor companies:
For us the key is to have great products, and that's why we're adding in this third wave of products [Kaby Lake]. At the first approximation, that's the best correlation of gross margin; when you've got leadership products, you tend to have a healthy gross margin.
Although doing the legwork to port the Cannonlake CPU core to 14-nanometer would be nontrivial, my view is that customers (particularly the system OEMs) would likely be willing to pay a premium for more CPU performance.
So, for example, let's assume that it costs Intel $100 million to port the Cannonlake CPU core to 14-nanometers. Then, let's suppose that because the system vendors find more a bit more value in 14-nanometer chips with the Cannonlake CPU core over ones with the Skylake core.
If the additional CPU performance that comes with the Cannonlake core delivers enough performance to warrant, say, a $2 per-chip premium over a part with just a higher-clocked Skylake, then across roughly 200 million PC units, Intel could potentially realize a four times return on its investment.
Now, unfortunately, these numbers are hypothetical. Perhaps in the real world, the analysis of the potential ROI yields a scenario that's less attractive.
However, if Intel's internal analysis shows that there is potentially a real and meaningful financial benefit to backporting Cannonlake to 14-nanometer for Kaby Lake (and the mere fact that Intel is even doing the Kaby Lake stopgap part in between Skylake and Cannonlake may suggest so), then there's a reasonable chance that the company might actually do it.