In a prior piece, I discussed the likely launch timeframe for Intel's (INTC +4.36%) upcoming 14-nanometer Atom-based Denverton micro-server processor. However, one question that I hadn't yet found the answer to was which 14-nanometer low-power Atom Intel will be using for Denverton.
At first, it would seem almost obvious that the successor to the 22-nanometer Silvermont-based Avoton would be powered by the successor to Silvermont, known as Airmont. However, as I noted previously. Intel did skip completely over Airmont for smartphones in a bid to accelerate the time-to-market for Airmont-successor Goldmont, so there was room for debate as to which 14-nanometer Atom core the Denverton server chip would use.
Today, thanks to new evidence, I can with a high degree of confidence state Denverton will sport Airmont, and not Goldmont, cores.
How do we know this?
Take a look at the following snippet from the LinkedIn profile of an Intel employee:
Source: LinkedIn.
The engineer claims to have worked on "the coherent and memory fabric for Intel's 22nm & 14nm Atom cores (Silvermont, Airmont) used in Intel phone, tablet, and u-server [micro-server] SoC." Since we know that Intel will be doing a 14-nanometer micro-server chip, and since the engineer specifically talked about Silvermont/Airmont (and didn't mention Goldmont), this is pretty strong evidence that Denverton will be based on Airmont.
What else can we deduce about Denverton?
From how Intel has generally positioned its Atom-based micro-server chips, we know that Denverton will be a "lower end" part relative to the Broadwell-based Xeon D that is slated to launch in the second/third quarter of 2015, according to CPU World.
We know that the fastest Denverton part will likely offer comparable or worse multi-core CPU performance relative to an 8-core Xeon D, which will feature Broadwell CPU cores. To try to figure how many Airmont cores Intel will likely put on this chip, it is helpful to understand the performance differential between an Intel "big" core and an Intel "small" core.
Here's the best Avoton (Atom C2750) available today, which is rated at a 20-watt thermal design power, compared to a current-generation 45-watt Haswell-based Xeon E3-1265L:
Source: Primate Labs.
Note that the per core performance of the Xeon E3 is over three times that of the C2750 part, but thanks to twice as many cores available to it, the multithreaded performance of the Xeon E3 is under twice that of the C2750.
The good news here is that the Xeon D family will, in the same power envelope, feature up to eight cores. This could mean that Intel is likely to be less aggressive in the per-core performance delivered, but I expect multicore performance to be significantly up -- nearly by a factor of two.
If Airmont is just a slightly better Silvermont at a higher clock speed, then Intel could probably comfortably double the number of cores in Denverton over Avoton (as it is doing with Xeon D relative to Xeon E3) and the Xeon D would still likely be quite a bit faster. Given the high core counts from some of the other microserver players, I'd be shocked if Intel didn't at least double the core count in Denverton over Avoton.
