I vividly remember then-Intel (NASDAQ: INTC) executive David Perlmutter claiming in 2013 that the chipmaker's newly announced Silvermont core was far superior to ARM's (NASDAQ: ARMH) Cortex A15.
"Cortex A15 is not even close to [Intel's] Silvermont. They are higher power and much behind us on performance which means they are on the wrong scale," Perlmutter told PCWorld.
How things have changed.
ARM's Cortex A72 does much better
In the time since the Silvermont release, Intel has launched a follow-on known as Airmont. Intel hasn't said much about the core, but published performance numbers for the Atom x7-8700 (which uses the Airmont core) suggest it is essentially Silvermont implemented in Intel's 14-nanometer technology.
The 14-nanometer process reduces the core's silicon footprint and should lead to better performance-per-watt, but peak performance doesn't really improve.
In contrast, the ARM Cortex A57 pushed peak performance forward relative to the A15. ARM's recently announced Cortex A72 is similarly said to push performance and power efficiency forward from the A57.
Given that the A15 and Silvermont offered fairly similar peak performance, I'd be willing to bet that the Cortex A72 delivers substantially more performance than Airmont (assuming Airmont on Intel 14nm and A72 on TSMC (NYSE: TSM) 16nm FinFET+).
Intel needs Goldmont ASAP
Intel originally said chips based on its new Atom core, code-named Goldmont, would arrive in mid-2015, but those chips were pushed out to 2016.
The good news for Intel is that the A72 isn't expected to debut in chips built on leading-edge manufacturing technology until sometime in next year as well. This means the A72's real competition isn't Airmont (which is on the market today), but Goldmont.
If Intel can deliver Goldmont-based high-end applications processors in early 2016, then the company has a shot of being competitive with CPU performance in modern ARM solutions in terms for the 2016 product cycle -- assuming Goldmont is a substantial leap from Silvermont.
The challenge for Intel in 2017
The follow-on core to the ARM Cortex A72 is known as Ares. The implication on ARM's part, according to an article in EE Times, is that the A72 was essentially an enhancement of the A57 focused on efficiency that was completed in "a little more than a year."
In that same article, an ARM representative reportedly said A72 borrowed some functional units from a "high-end core which was a brand new redesign."
If Intel follows its typical development cadence, the follow-on to Goldmont should basically be a slight enhancement implemented in the company's upcoming 10-nanometer manufacturing technology. In that case, it might be tough for Intel to compete architecturally with ARM's "brand new" core.
Intel needs to update investors on its strategy here
At this point, I think Intel needs to update investors regarding its strategy on the mobile core development front. Right now, it seems ARM -- and its partners -- are moving much faster than is Intel.
To illustrate why this is so surprising, note that ARM's entire research and development budget was about $260 million last year; Intel's came in at over $11.5 billion.
Intel should be able to make meaningful advances in its low-power processor cores each and every year. For the sake of Intel and its stockholders, I hope it plans to do so going forward, because "advances" like we saw with the Silvermont to Airmont transition just aren't enough to stay competitive.
