A few days ago, mobile chipmaker Qualcomm (QCOM -1.14%) announced its first processor aimed at cloud computing applications. Qualcomm’s new chip likely represents the first serious competition from an ARM architecture-based processor that chip giant Intel (INTC 0.61%), which dominates the market for datacenter processors, has ever faced.
Qualcomm’s new chip, sold under the Centriq 2400-series branding, is manufactured using Samsung’s (NASDAQOTH: SSNLF) 10nm LPE technology.
Image source: Intel.
One of the defining characteristics of a chip manufacturing technology is how densely packed the individual elements of a chip, known as transistors, can be. The higher the packing density, the more features and functionality can be built in a certain physical footprint.
Intel has long claimed that it has a significant lead in the density of its technologies compared to what competing chip manufacturers, like Samsung, can deliver. I’ve long challenged this claim, particularly as Intel’s main competitors are both currently building mobile processors on 10nm technologies that are denser than Intel’s currently-available 14nm technology.
If those contract chipmakers were only building mobile processors with their technologies, rather than products that directly competed with Intel’s PC processors and datacenter chips, Intel could brush the situation off as trying to compare apples with oranges.
Intel can’t hide behind that excuse anymore.
Density leadership lost in servers
According to Intel’s own metric for measuring the density capabilities of various manufacturing technologies, Samsung’s 10nm LPE technology can stuff about 50 million transistors into a square millimeter. Intel’s same metric says that its currently in production 14nm technology can cram 37.5 million transistors into a square millimeter.
This means that Qualcomm is now shipping datacenter chips built using a technology that, by Intel’s admission, is denser than what Intel’s current best data center chips are built using.
To make matters worse, Intel doesn’t have a chance of fixing this issue for at least another generation. Intel’s next-generation datacenter processors, known as Cascade Lake, will be manufactured using a performance-enhanced version of the 14nm technology that’s used to build the company’s current Skylake datacenter processors.
Intel is likely to transition its datacenter processors to some variant of its 10nm technology (I believe it will be 10nm+ for the first 10nm Intel datacenter processor) in 2019, which should buy the company a large boost in density. Intel claims that its 10nm technology has an intrinsic density of around 100 million transistors per millimeter -- a near tripling of density.
Image source: Intel.
While Intel’s 10nm technology is substantially denser than competing 10nm technologies, the reality is that by the time Intel brings its 10nm technology into the datacenter, it’ll have to face off against chips built using foundry 7nm technologies.
The good news is that Intel’s 10nm technology will likely be roughly on par with the various foundry 7nm technologies that it’ll ultimately compete with (in fact, Intel’s 10nm technology may even be slightly denser).
The bad news for Intel is that it’ll still have gone from a clear leadership position in density at the beginning of its 14nm generation to essentially parity with its key competitors in 2019 and beyond.
Intel's weakened competitive positioning with respect to chip manufacturing is due largely to the fact that the company's 10nm technology has been delayed by a matter of years.
There’s, of course, more to chip manufacturing technology than just pure density (performance and power efficiency are big ones), but considering how significantly Intel has played up its claimed density advantage over the competition, falling behind Qualcomm in this metric is nothing short of disappointing.
