May 05, 2016
The Intel Atom architecture is surprisingly confusing and complicated given the breadth of chips offered, and with Intel's announcement that it is axing some atom chips from their line-up many people are equally confused that this means for other OEMs product pipelines. There are two products specific products that people are worried about, The Surface 4 and the Surface Phone. Intel just has too many SKUs with random missing features which makes is a quagmire to navigate.
We have heard of Intel axing various chips before, even Pentium and Celeron whose brands still confusingly live on. Why are these not Core i3, or Core i1 branded given their architecture? Like the Core architecture which has chips from 6 W for 2 in 1s up to 130 W TDP for servers, Atom has chips from 2 W phone SoCs to 20 W server chips.
It is widely being reported that Intel is Axing Atom and confusingly the media is then talking about SoFIA, and Broxton chips, and X5 Cherry Trail. I will address each of these individually.
Avoton, based on the media release, Atom in the data centre isn't going anywhere. Denverton, the successor to Avoton has not been mentioned in the culling and are still expected to ship this year. These are perfect for small NAS boxes and network appliances which don't need all the bells and whistles attached to the much more expensive Xeon D chips.
SoFIA is the mobile system on a chip (SoC) which includes LTE radios. There was so little interest in x86 mobile phones that I'm not sure what Intel was thinking. Intel has dabbled in mobile before with the X Scale ARM chips in the early 2000s that Intel later sold, whoops. This is despite X scale having a lead in the field and used in a majority of Windows Mobile phones. I'm not sure Intel understands mobile phones as this is their third exit from the sector after the abandoning of the MeeGo operating system. We can assume Intel has exited mobile phones for good.
Broxton is the next generation tablet SoCs and includes all Goldmont SoCs including Morganfield and Willow Trail. Willow Trail's cancellation is the major concern for Windows tablets as it is the successor to Cherry Trail which are used in the Surface 3. These Atom X5 and X7 chips were relatively cheap for x86 costing between $20 and $35 USD. The problem for Intel is two fold, these chips aren't quite as powerful as the ARM chips used by Apple in specific artificial benchmarks and various specific multimedia workloads. But if Intel made Atom much faster they would start to erode into the space the Core M processors sit which are priced an order of magnitude higher between $281 and $393 USD.
At 4.5 W TDP these Core M processors make sense for Intel's strategy where they intend to extract more of the value from the tablet and 2 in 1 manufacturers and drive the market towards higher priced products. But at $281 a Core M3 would be almost half the component cost of a premium tablet.
Enter Apollo Lake, the successor to the little known Braswell which will be launched beside Kaby Lake. The branding of the current Braswell chips is a mess with Atom, Celeron, and Pentium brands all thrown in the mix. But with a maximum price of $161 these start to look like the Atom successor Intel really needs in a 6 W TDP. In fact looking at the passmark benchmark per dollar, you can see that they fall into the traditional Intel pricing based on performance. Cherry Trail by comparison was exceptional value providing either high profits for OEMs like Microsoft or cheap bottom of the rung no name brand tablets.
But despite the higher TDP there is still a performance problem. The 1.6 GHz Atom X7 Z8700 at $35 has a pass mark score of 1956 vs the Braswell 1.6 GHz Pentium N3700 at $161 with a pass mark score of 1902. That is performance is the same and performance per dollar and performance per Watt is considerable worse.
When you realise that despite the branding Braswell are Atom cores and you can see significant overlap in Intel's product line-up. Intel is not giving up on Atom, they are consolidating their efforts to go after high end ARM chipsets in the 4-6 W category and giving up on the lower price point 2 W category. These low powered Atoms have been lagging higher powered ARM chips leaving Intel a bit red in the face. While I expect Apollo Lake to deliver a modest performance bump and a power reduction inline with Skylake putting it head on with the power envelope of Cherry Trail. Intel still won't want to erode the performance of the higher priced core M chips. Within the next 24 months low priced x86 tablets are officially dead with Intel overnight adding up to $126 to BOM costs for the processor and reducing other BOM component costs by about $5. Intel's slice of the pie is getting bigger.
The major improvement still needs to be the integrated graphics that will be upgraded to gen 9 as on Skylake with performance yet to be announced. I would expect a serious bump to directly take on Apple's A9X with PowerVR 7XT which dominates even the Surface Pro 4's HD 520 and 24 Execution Units (EU) which Apollo Lake is expected to have. The inclusion of hardware decoding for Google's VP9 is also very welcome especially with the inclusion with the upcoming redstone version of Edge for YouTube playback.
Apollo Lake's launch in the second half of 2016 is intended to smoke ARM pushing OEMs to produce higher performant and more pricey 2 in 1s and increasing Intel's slice of the pie. Apollo Lake's support for USB3, PCIe, SATA, DDR 4, and Gen 9 Intel HD graphics give Intel an edge that Cherry Trail never could. It is likely the success of the Atom X7. The inclusion of SATA3 and M.2 x4 PCIe lanes is important, despite the promise that eMMC 5.0 can do 400 MB/s vs SATA 6 Gb/s having a real world performance wall of 550 MB/s after overheads. On the memory even Cherry Trail supports up to 8 GB of RAM which would make Windows supporters jump for joy had a premium OEM exercised this option. Windows supports have been fairly critical of Microsoft's decision to include a 2 GB Surface 3 and more lately for Apple to have released a 2 GB iPad Pro.
We can conclude that Intel still has an Atom processor in the works based on Apollo Lake that is suitable for a Microsoft Surface 4, it will be soon, and it has features that could address some performance complaints about the Surface 3, specifically storage speeds. How these processors are to be branded is yet to be seen, but Intel needs to bring something new to the table to demand the higher prices they will be asking. The Atom X7 rebrand did a lot to showcase Atom as an acceptably high performant tablet CPU, but they will need to bring something newer to the table than recycling the Celeron and Pentium brands.