Microsoft’s Disappointing Surface Go Reportedly Used x86 at Intel’s Insistence

Microsoft’s Disappointing Surface Go Reportedly Used x86 at Intel’s Insistence
Microsoft’s Disappointing Surface Go Reportedly Used x86 at Intel’s Insistence

According to Paul Thurrott, many of these issues can be traced to Microsoft’s decision to use an x86 CPU rather than its original plan for a Qualcomm Snapdragon. He writes:

In addition to needing to meet a certain price point for Surface Go to make sense, only the Pentium Gold chipset matched Microsoft’s thermal requirements given the form factor. For Surface Go to be this thin and light, and fanless and silent, Pentium Gold was literally its only viable option in the Intel stable.

Performance tests on Surface Go by PC Mag.
Performance tests on Surface Go by PC Mag.

Thurrott doesn’t go on to praise Windows on ARM as the obvious solution Microsoft ought to have embraced. In fact, according to him, it’s the wrong product for an entirely different set of reasons. Where the Pentium Gold fails to deliver on the performance and battery life that users want, Windows on ARM offers abysmal overall performance on x86 code, only supports 32-bit applications, and has a very limited set of hardware drivers compared to x86. Snapdragon 850 should improve the situation a little, but it won’t really get much better until Snapdragon 1000, which might appear on store shelves next spring if all goes well. With as much as a 2x performance boost expected from that platform, the Snapdragon 1K might be the part that makes Windows on ARM worth taking seriously.

But this presumes an answer to the question: “Would Microsoft allow Intel to bully it into releasing a piece of hardware on a different platform than it originally intended?”

The answer: “Yes, Absolutely.” And we can prove it.

Remember Vista Capable?

Nearly a decade ago, Microsoft was sued over its decision to label some laptops that weren’t capable of running the full Vista accelerated UI as “Vista Capable.” Users bought these machines, brought them home, and discovered they didn’t run the OS particularly well or with much of its eye candy enabled. They sued. As a result of that lawsuit, nearly 200 pages of private, internal Microsoft e-mail were made public.

I covered that lawsuit for Ars Technica at the time and reading those documents was an eye-opening experience into how these decisions are made at a huge company like Microsoft. At one point, Microsoft had represented to HP at a very high level that it was committed to only supporting Windows Vista on chipsets that could handle the new WVDDM (Windows Vista Display Driver Model). Only these systems would be badged as “Vista Capable.” HP, as a result, committed to buying only expensive chipsets from Intel that could perform these tasks. But Intel was extremely unhappy with Microsoft’s decision. Only selling Vista on WVDDM-capable laptops meant that most of Intel’s existing chipsets would be unable to support the OS. This decision was worth a great deal of money to Intel and the company hammered Microsoft on the point.

Under pressure, Microsoft flipped. Despite the fact that companies like HP had committed to only buying Intel’s more-expensive chipset lines for all their laptops on the belief that only these chipsets would be supported for Vista, Microsoft changed its own guidance and expanded the Vista Capable program for laptops that could support some (but not all) of Vista’s features, approximately six months after representing to HP that it absolutely would not buckle on this point. And the company admitted its own reasoning, internally.

On February 26, 2007, Microsoft general manager John Kalkman sent an email to his team discussing how the “Vista Capable” definition would be changed going forward. Kalkman wrote:

In the end, we lowered the requirement to help Intel make their quarterly earnings so they could continue to sell motherboards with 915 graphics embedded. This in turn did two things: 1. Decreased focus of OEM’s planning and shipping higher end graphics for Vista ready programs and 2. Reduced the focus by IHV’s to ready great WHQL qualified drivers. We can see this today with Intel’s inability to ship a compelling full featured 945 graphics driver for Windows Vista.

That was 2007, not 2018, and some of the players are different — Jim Allchin, who actually handled the HP meetings and promised them that MS would only support Vista on high-end chipsets, left the company after Vista was released. But as to whether Microsoft would historically change its product plans to favor a solution preferred by Intel? That’s not even a question. Microsoft tore up its entire launch strategy for its first consumer OS launch in five years (remember Windows XP had an unusually long life) to help Intel make its quarterly numbers, and it broke high-level commitments to HP and other OEMs to make that happen.

It’s not clear that the Snapdragon 835 would have actually resulted in a better Surface Go. Given its weak performance in x86 apps and Microsoft’s undoubted desire to stay away from a second Surface RT debacle, x86 may have always been the better choice. But the ongoing difficulty of finding a low-cost tablet solution that can pack both acceptable performance and decent battery life is an example of how it’s increasingly difficult to hit our desired goals in semiconductor design without compromising some aspect of the overall experience. Snapdragon 850 and 1000 may prove useful towards this goal in 2019, but chances are the progress will be incremental — the old days of free and easy improvements are now long behind us.