At the Intel(INTC 3.75%) investor meeting back in 2013, CEO Brian Krzanich said the company wanted to make its Atom family of low-power processors equal partners to its Core family of processors. That did not mean Intel planned to make its Atom and Core processors perform equally (they are targeted at different power envelopes), but it was taken to mean that Intel would treat both of them with equal importance to the company.
In fact, over the long haul, while the big Core line is still likely to bring in massive profits for Intel, I can see Atom doing very well for the company, too. Intel has already used its Silvermont Atom core to great effect in low cost PCs, microservers, and tablets. Had they launched in 2013, when Silvermont first launched, the phone chips based on Silvermont could have offered competitive performance for fairly high-end devices as well.
It is not hard to see how Atom-based processors are going to be important to the Intel business over the long-term. However, there is one way the Atom family of processors is still not quite equal to its Core siblings.
How many teams work on Atom, again?
Readers might be familiar with the "tick-tock" strategy. In any given year, Intel is either refining a previous architecture and building it on new manufacturing technology or bringing a new architecture to a proven technology.
So, for example, the fifth-generation Core, known as Broadwell, is a "tick" because it represents a refinement of the Haswell architecture on the new 14-nanometer manufacturing technology. Skylake, due in the second half of this year, is a "tock" -- a new architecture built on the same 14-nanometer technology used to build Broadwell.
It is well known that Intel has two parallel CPU core teams. It has a group in Hillsboro, Oregon, known as the Converged Core Development Organization, or CCDO, and it has a group in Haifa, Israel, known as the Israel Design Center or IDC. Each team alternates responsibility for tick-tock pairs.
For example, the Hillsboro team was responsible for Haswell and Broadwell, while the Haifa team was responsible for Sandy Bridge and Ivy Bridge. It is expected that Skylake and Cannonlake will be done primarily by the Haifa team once again.
On the Atom side of things, Intel has a team in Austin, Texas, and this team is responsible for Atom core development. According to Intel, it does not have a parallel site working on Atom core development similar to "big" cores.
The implications
During a presentation in 2013, Intel said that it planned to refresh its Atom cores yearly from there on out to be more competitive in the market. However, preliminary performance results show that Airmont, the follow-on to Silvermont, does not bring a significant performance increase over Silvermont. Goldmont, however, is expected to bring a sizable increase in performance over both.
The question, though, is whether this will be enough to keep Intel competitive in the mobile market. For example, ARM Holdings (ARMH +0.00%), which supplies processor designs to system-on-chip designers, has quite rapidly iterated its processors.
To illustrate, the first Silvermont-based chips hit the market in September 2013, and the first Airmont-based chips will hit the market in the first half of 2015 (let's call it May). In the following table, I show the first commercial availability of chips with the indicated ARM cores:
|
CPU core |
Cortex A12 |
Cortex A17 |
Cortex A57 |
Cortex A72 |
|---|---|---|---|---|
|
First availability |
2H 2014 |
1H 2015 |
2H 2014 |
2H 2015 |
Source: AnandTech
In the time it has taken Intel to move from Silvermont to Airmont, ARM and its partners have launched chips based on the Cortex A12, A17, and A57. Cortex A72 chips from MediaTek and Qualcomm should be coming this year as well.
Is this a processor IP problem or a chip integration problem?
To be perfectly fair to Intel, the company had expected Airmont-based chips to hit the market in late 2014, and Goldmont-based chips to show up in mid-to-late 2015. However, delays at the chip level (not necessarily the CPU core development level) pushed those products into the first half of 2015 and 2016, respectively.
However, I do wonder if Intel had parallel CPU core development teams, it could make bigger strides per generation. The performance delta between the Cortex A15 and the A57 were relatively sizable, and if ARM's claims are to be believed, the jump from A57 to A72 should also be quite large. In contrast, Airmont was a small improvement over Silvermont, and it remains to be seen how big of a jump Goldmont is.
If Intel can start making more meaningful jumps with its yearly CPU core introductions, then the company can probably get along fine with a single Atom CPU core development team. If Intel has to wait for every other CPU core to get significant improvements, then perhaps adding a second parallel Atom development team would be a good idea.
