Intel Kills the Atom SoC Powering Microsoft’s HoloLens

Last month, we wrote about new information about the follow-up to Microsoft’s HoloLens, and how the new device would integrate an AI co-processor. At the time, we speculated that Intel’s Atom SoCs might not be long for this world, though we didn’t expect the hammer to drop quite as quickly as it has. Intel is killing its Atom X5-Z8100P processor, the chip that powers the Microsoft HoloLens.

This chip isn’t publicly listed on Intel’s website (Ark.Intel.com; useful for all CPU identification tasks), but has been identified (PDF, but with security warning)  as a Cherry Trail-based part clocked at 1.04GHz. Intel’s decision to cancel its Atom-based SoCs as it exited the smartphone and tablet business has left Microsoft without a follow-up product, though the fact that Intel is exiting the market for a chip that may only have had one customer implies Microsoft and Intel have worked this issue out amongst themselves.

Another reason to view the situation as a mutual agreement is Intel hasn’t killed its other Cherry Trail Atom SoCs in the same family:

Plenty of Intel Cherry Trail SoCs are still on the market.

This raises the question, however: If Intel isn’t providing a CPU solution for HoloLens 2 (our title), who is? We may not know which company, specifically, is building the solution, but there are only three options, and only one of them makes sense:

1). Microsoft has paid Intel to build it a semicustom Atom SoC based on Goldmont.
2). Microsoft paid AMD to build it a semicustom SoC based on a 14nm Jaguar core.
3). Microsoft contracted with an ARM vendor.

#3 is the only realistic option here. Microsoft may have paid Intel for a custom Atom SoC before, but that was back when Intel was still committed to selling Atom chips across the tablet and smartphone markets, including in Microsoft’s own Surface 3. It’s one thing to be a semicustom customer for the mildly custom part that’s being sold elsewhere, and something else entirely to be the only customer the company expects to have for an entire hardware line. MS can scarcely hope to make its investment back in volume; there’s no indication that the company is planning a broad consumer rollout for HoloLens 2.

The second option doesn’t make sense either, for the same reason. We know AMD has ported Jaguar to 14nm, but the SoC designs for the PlayStation Pro and Xbox One X don’t have a mobile-compatible GPU bolted on. And AMD has made only the faintest of overtures towards the tablet market since it sold the GPU IP that became Qualcomm’s Adreno nearly ten years ago. Again, Microsoft isn’t going to want to foot the entire bill for a low-cost x86 SoC and associated graphics hardware, especially when AMD, like Intel, has no plans to compete in the tablet or smartphone space.

That leaves option #3. Here, there are a plethora of options and plenty of companies who would love a high-profile win in a device like HoloLens. Both Qualcomm and Imagination Technologies have GPUs that could handle the processing, especially since Microsoft is beefing up the role and capability of its artificial intelligence processing.

This last point is presumably why the firm wouldn’t turn to a company like Nvidia. While Nvidia might have some silicon in its arsenal that could be repurposed for this role, strapping Microsoft’s custom AI silicon into the same hardware that features Nvidia’s AI silicon (via the GPU) would be redundant. Besides, Nvidia has gotten out of tablets and smartphones as well, and is focused on vehicles, which have much more robust cooling options and room for larger boards than your average headset.

Barring a Microsoft announcement that it’s getting into the CPU business, ARM is the only likely fit here.