Image source: Intel.
Microprocessor giant Intel (INTC 0.99%) has built something incredible with its Data Center Group. This business generated approximately $16 billion in revenue and a whopping $7.844 billion in operating income last year. Although the financial results of the segment speak for themselves, there's still room for the company to improve its execution.
A quicker, more consistent product-release cadence
Intel has multiple server product families, but its "bread and butter" processor family is the Xeon E5 series. These processors can be used in single, dual, and quad processor configurations.
At any rate, the release cadence for these products has, at times, been inconsistent:
|
CPUs |
Launch |
|---|---|
|
Nehalem-EP |
1H 2009 |
|
Westmere-EP |
1H 2010 |
|
Sandy Bridge-EP |
1H 2012 |
|
Ivy Bridge-EP |
2H 2013 |
|
Haswell-EP |
2H 2014 |
|
Broadwell-EP |
1H 2016 |
|
Skylake-EP |
1H 2017 (projected) |
Source: Intel ARK database, BenchLife.info.
As the table above shows, there was a two-year gap between Westmere-EP and Sandy Bridge-EP, a year-and-a-half gap between Sandy Bridge-EP and Ivy Bridge-EP, and a year-and-a-half gap between Haswell-EP and Broadwell-EP.
It looks as though Intel will get back on track (in terms of a one-year product-release cadence) as it transitions from Broadwell-EP to Skylake-EP.
That being said, not much has leaked to the web about when we should expect the company to transition to its next-generation server processor family, known as Cannonlake-EP.
10-nanometer health: key to Cannonlake-EP
Although the issues that Intel seemed to have in going from Westmere-EP to Sandy Bridge-EP were clearly not driven by manufacturing technology (Westmere-EP and Sandy Bridge-EP were built on the same 32-nanometer technology), the latest slips in schedule do seem to have been.
In going from Sandy Bridge-EP to Ivy Bridge-EP, Intel was managing the transition from its 32-nanometer technology to its 22-nanometer FinFET (Fin Field-Effect Transistor) manufacturing process. Likewise, in going from Haswell-EP to Broadwell-EP, Intel was dealing with a move from its 22-nanometer technology to its 14-nanometer process technology (which had some widely publicized yield issues).
In going from Skylake-EP to Cannonlake-EP, Intel will once again have to deal with a transition to a new manufacturing technology. This time, the company will have to go from its 14-nanometer process to its new 10-nanometer technology.
Intel was originally supposed to go into high-volume manufacturing on its 10-nanometer process at the end of 2015, but now the first Intel products based on this technology aren't expected to arrive until late 2017.
If the 10-nanometer technology is in great shape by the end of 2017, then I could see the company going into production on Cannonlake-EP at that point, for launch in the first half of 2018. If the process is in rough shape, then Intel will face a difficult choice.
If competition in the server market is fierce, and if 10-nanometer yields are not great, Intel may choose to go into production on 10-nanometer server parts anyway, and take the cost structure hit in order to maintain a competitive edge in the marketplace.
If competition isn't as fierce, but 10-nanometer yields still aren't great, Intel may ultimately opt to push Cannonlake-EP into the second half of 2018 to try to protect its margins.
Why Intel shouldn't push Cannonlake-EP out
The problem with pushing out a new server chip is that it has repercussions on future platforms. Such delays give competitors -- and there are a lot of companies vying for a piece of Intel's lucrative server pie -- greater opportunity to strike.
In my view, Intel needs to make sure that it sticks to a one-year product release cadence in servers. It's much better for Intel to take a short-term hit to gross profit margins due to relatively poor yields, rather than to ultimately see a gross margin drop because aggressive competitors have driven pricing down.
