At its investor meeting back in February, Intel (INTC +0.30%) revealed that it would be putting its data-center focused chips on new manufacturing technologies first.
This represents a change from how Intel had done things in prior years. Under the old product development methodology, Intel would debut its newest chip manufacturing technologies on its personal computer processors, and then at some point afterward, release data-center products built on that technology.
Image source: Intel.
For example, Intel launched its high-volume 14-nanometer Broadwell-U parts in the first quarter of 2015, but its Broadwell-EP/EX processors for data centers didn't launch until the second quarter of 2016.
Similar gaps still occur today: The first data center processors built on the company's 14-nanometer+ technology launched earlier this month, about a year after the first personal computer processors built on that technology. And, very shortly, Intel's personal computer chips are set to transition to the company's newer, more efficient 14-nanometer++ technology.
How data center chips will jump ahead
In this column, I'd like to explain the path that Intel is likely to take to get its data center products first to new manufacturing technologies.
The maneuver should be a fairly straightforward leapfrog: Data center will skip the first 10-nanometer iteration.
Intel has indicated that data center chips will be the first products to utilize the company's upcoming 7-nanometer technology -- which should go into volume production in either 2020 or 2021. Further, then-data center general manager Diane Bryant said at the company's investor meeting earlier this year that the first products to use Intel's 10-nanometer++ technology would also be data center chips.
We know that Intel's personal computer products should transition to 14-nanometer++ later this year (ahead of the transition for Intel's data-center parts, which should happen in 2018), and we know that the company intends to launch the first 10-nanometer personal computer chips sometime in the middle of 2018.
So, for Intel to achieve its desired goal, its data-center road map should look something like this:
|
2017 |
2018 |
2019 |
2020 |
2021 |
|---|---|---|---|---|
|
Skylake-SP (14nm+) |
Cascade Lake (14nm++) |
Ice Lake-SP (10nm+) |
Sapphire Rapids (10nm++) |
Future (7nm) |
Intel's personal computer chip road map would, correspondingly, look something like this:
|
2017 |
2018 |
2019 |
2020 |
2021 |
|---|---|---|---|---|
|
Kaby Lake-U Refresh/Coffee Lake (14nm++) |
Cannon Lake (10nm) |
Ice Lake (10nm+) |
Tiger Lake (10nm++) |
Future (7nm) |
You'll note here that the personal computer processors will still be a generation ahead in 2018, but in 2019, the data center and personal computer parts should largely converge in terms of manufacturing technology (though I expect Ice Lake for personal computers to launch earlier in the year than Ice Lake-SP for data center).
Then, in 2020, the 10-nanometer++ data-center parts could take a decisive lead by launching earlier in the year than the 10-nanometer++ personal computer parts. That pattern could then extend to the first iteration of the 7-nanometer technology and beyond.
The launch gap between Intel's personal computer processors and its data center processors wouldn't be huge -- it could be a matter of just a few months -- but Intel appears to have a clear path to putting its data center chips first, if it can execute properly.
