Several years ago, Intel (NASDAQ:INTC) disclosed its launch plans for low-power, highly integrated system-on-a-chip designs aimed at the server market. These chips, unlike the company's workhorse Xeon processors, are based on Intel's Atom architecture.
These products don't offer the kind of performance one would expect from Xeon processors, but they are intended to be lower cost and, for some workloads, more efficient solutions than the Xeons.
In late 2013, Intel released its second-generation Atom-based micro-server part known as Avoton that, to this day, compares very favorably in terms of performance and power efficiency compared to solutions in the market based on ARM Holdings' designs.
Thanks to CPU World, we now know a lot more about the successor to the Avoton part, known as Denverton -- enough to determine how competitive it might be in the marketplace.
Huge improvements across the board
According to CPU World, Denverton will include 16 next-generation CPU cores, a doubling in core count from the prior-generation Avoton part. Additionally, these cores, as we learned a little while ago, will be Intel's Goldmont CPU cores, representing a two-generation leap from the Silvermont CPU cores found inside of the Avoton part.
CPU World says each pair of CPU cores -- known as modules -- will feature two megabytes of level two cache. Since prior leaks suggest consumer-oriented versions of Goldmont will feature one megabyte of level two cache per core pair, it would seem the variant of Goldmont inside Denverton may be specifically tuned/configured for server workloads.
The new chips are also said to include support for up to 128 gigabytes of DDR4-2400 memory, up nicely from support for 64 gigabytes of slower DDR3-1600 memory in Avoton.
Finally, CPU World says the Denverton chips will see big improvements in integrated I/O capability (quite important for the kinds of workloads people generally associate micro-servers with). Updates here include support for 10Gb Ethernet (up from support for 2.5Gb Ethernet) as well as 16 lanes of PCI-Express 3.0 connectivity (Avoton featured 16 lanes of PCI-Express 2.0 connectivity).
Coming in the second half of 2016, will it be competitive?
According to CPU World, Denverton will arrive in the second half of 2016, suggesting no less than a three-year delta between Avoton and Denverton. Given how much Intel invests in its data center group, and given that these investments seem to be increasing, it would seem Intel doesn't place much priority on Atom-based server chips.
Even though Atom-based server chips appear to be relatively low priority within Intel, I still think Denverton will represent a solid, low-cost/low-power server chip offering when it arrives in the marketplace.
I believe that against ARM-based competitors in this market such as Applied Micro (NASDAQ:AMCC) and Cavium Networks (NASDAQ:CAVM) in the late 2016 timeframe, Intel will enjoy a multi-generation manufacturing technology advantage, which should translate into a material power efficiency advantage. I don't expect either Applied Micro or Cavium to ship products built on foundry 14/16-nanometer FinFET processes until at least 2017.
Qualcomm (NASDAQ:QCOM) has been public about its intentions to enter the server market as well, and I would expect its first products to arrive around late next year since it is sampling chips now. Unlike with Applied Micro/Cavium, Qualcomm's first server parts should be built on a foundry 14/16-nanometer finFET process, substantially negating Intel's manufacturing lead in this area.
Nevertheless, I suspect Qualcomm will be targeting Intel's higher-end dense server parts, known as Xeon D, with its first server processors rather than the upcoming Denverton part.
The future for Atom-based server chips?
I think Intel will continue to roll out Atom-based server chips going forward, if only in an attempt to leave no stone unturned. However, I don't think the pace of innovation here will be as rapid as it is in the company's higher-end Xeon E5/E7 and Xeon D product lines.
My expectation is that Intel will launch one Atom-based server processor per manufacturing technology generation, and that release will align with the "tock" processor at that node. If I'm right, we should see the successor to Denverton in two or three years.