Microsoft's long-running partnership with Intel has been growing wrinkles. The company built an emulator into its Windows 10 operating system so it could run applications built for architectures other than the x86 pioneered by Intel. And on Wednesday, it complicated Intel’s future in its growing cloud business.
Microsoft, which is based in Redmond, Wash., said that it would run several internal cloud services with chips based on ARM designs, which are also used in the vast majority of smartphones and inside gadgets like thermostats and sensors embedded in manufacturing plants.
The move amounts to one of the most significant endorsements of ARM servers, showing that the architecture has the performance and standardized code base to potentially encroach on Intel’s dominant position in the market for server chips, analysts said. The other large cloud computing companies, Google and Amazon, plan to use Intel's next-generation Skylake chips.
Microsoft reworked its Window Server operating system to run over the ARM servers, which will be used internally for applications in search, storage, databases, and machine learning. The announcement came at the Open Compute Platform Summit in Santa Clara, Calif.
“We feel ARM servers represent a real opportunity and some Microsoft cloud services already have future deployment plans on ARM servers,” said Leendert van Doorn, a distinguished engineer at Azure, Microsoft’s cloud computing unit, in a blog post.
On Wednesday, Microsoft also signed design partnerships with Intel, Advanced Micro Devices, and Nvidia, underlining how many different types of chips will ultimately live together in data centers.
Qualcomm and Cavium both devised Windows Server hardware based on ARM technology. Qualcomm built a server using its Centriq 2400 server chip with 10 nanometer technology and 48 computing cores. Qualcomm is giving out a small number of samples for testing, but the chip will start selling later this year.
It has been bumpy road to legitimacy for ARM server chips. Since first appearing five years ago, sales of the chips from companies like AMD, Applied Micro, and Broadcom have fallen flat. The hardware has struggled to match the standard code base and performance offered by Intel processors, analysts said.
Paul Teich, a principal analyst with Tirias Research, said that Microsoft’s announcement shows that the latest batch of ARM server chips were able to meet those conditions. The move, he added, also shows how the industry is craving more competition – an idea echoed by ARM itself.
“Faced with different requirements for workload optimization, data center architects have been calling for increased choice and flexibility in server architectures,” ARM said in a statement. “There is more opportunity now to increase dynamic competition in the server market.”
ARM chip makers are still facing an uphill battle to loosen Intel’s grip on the server market. The research firm IDC estimates that the x86 instruction set had been used in over 99% of all servers sold in the first three quarters of 2016. Intel reaped $17.2 billion in revenue from its data center business last year.
One stumbling block is that software written for x86 chips must be reworked for ARM hardware. But Microsoft’s van Doorn said that the ability to carefully match hardware and software would pay off. He said it is “economically feasible to optimize the hardware to the workload instead of the other way around.”