Intel Corp (INTC)

[wbmw]

This is definitely not what I envisioned happening when Silvermont was demoed at Computex in 2013.

Me neither, but does it matter, anymore? Rather than becoming a write-off, I'd rather see Silvermont get them a big piece of the mid-range and entry markets.

And frankly I don't think much of its chances at the low end against the ARM A53 which is a quarter of its size but has most of its performance at the lower clock rates in 28nm.

You're getting ahead of yourself. Phones based on MT6752 and MT6732 have just recently arrived. They are octa-core and quad core versions using Cortex-A53, and as far as I can tell by Googling, only a half dozen phones have been announced between both SOC's, and there are still no reviews or real information to be seen anywhere.

If you're using Geekbench as a proxy, then you're getting the inflatable benefit of the SHA and AES logic, which improved the scores by nearly 20x.

http://www.anandtech.com/print/8718/the-samsung-galaxy-note-4-exynos-review

Even so, Silvermont seems to be a better product, core for core. Discounting for SHA/AES tests, Silvermont seems to have a sizable lead.

http://browser.primatelabs.com/geekbench3/compare/1966035?baseline=1972455

As for die area, A53 on 28nm should be around ~5.5mm2 per cluster, if you extrapolate based on the data in the Anandtech review. An octa-core SOC would therefore have about ~11mm2 die area for the CPUs, on 28nm. Mediatek won't be moving to smaller process nodes in 2015, and it may take quite a long time before their lower-end designs make it there.

Meanwhile, you presume that Silvermont is 4x larger, which would put it at 11mm2 per dual-core cluster, and ~22mm2 for quad core. I don't have the data to dispute it, but based on the die photos, I think that's overestimating. Either way, it is larger, and it comes at the expense of margin, but not at the expense of the sale. I think Silvermont has the performance to win over Cortex-A53, and Intel has the right part for 2015 to compete in the lower tiers.

Based on the upper tiers, it will be a dog fight, no doubt. Mediatek will have an edge in MT performance, while Intel will have an edge in ST. Again, correcting for AES and SHA should give you an idea of how other benchmarks will perform.

http://browser.primatelabs.com/geekbench3/compare/1977307?baseline=1972455

Besides, I think Intel gets an edge overall in terms of their marketing, their sales channels, and their customer relationships. If they can get close to Mediatek in capabilities, I think they'll earn a chunk of MSS easily. The first 20-40% is a given.

After that, they need to pull ahead. Moving the design to 14nm in their own fabs should be part of the strategy. They'll be on an advanced process well ahead of Mediatek and other low-end vendors. I think by 2017, Intel should have a >40% footprint in the entry segments of the market, and >20% in the mid-range. They could be shipping a couple hundred million phone SOC's in 2017.