When Linus Torvalds thinks about his favorite chip architecture, he thinks about the x86 processors from Intel and Advanced Micro Devices.
What Torvalds doesn’t think about is the ARM architecture, a point he made this month during a talk at the annual conference hosted by Linaro—an organization that is dedicated to developing a Linux ecosystem around ARM—and in front of a crowd of ARM developers.
During a question-and-answer session with Linaro CTO and former ARM employee David Rusling at the Linaro Connect 2016 show, the Linux creator was asked to name his favorite chip architecture. He pointed to x86, saying it was more open and more unified, and more a broader ecosystem than the ARM architecture.
“x86 is still the one I favor most, and the reason for that is the PC,” Torvalds said. “Not because of the architecture; it’s because of the infrastructure. It’s because it’s there and it’s open and it’s open in a way no other architecture is. … It turns out the instruction set and the core of the CPU is not very important. It’s one of those big differentiating factors that people kind of fixate on, but it really doesn’t matter that much in the end. What matters is all the infrastructure around the instruction set, and x86 right now has all that infrastructure.”
The infrastructure, ecosystem and compatibility that x86 has over ARM is important, he said. ARM is the dominant architecture in the mobile device space; almost every smartphone and tablet is powered by low-power systems-on-a-chip (SoCs) that are designed by ARM and manufactured by such companies as Qualcomm and Samsung.
However, it’s the way that the ARM architecture is used that makes it difficult for hardware vendors, Torvalds said. ARM designs the CPU architecture and licenses it to other companies, which put their own technologies on top of it and then sell those chips to device makers. Because of this, Google can’t offer a single build of its Android mobile operating system—which is based on Linux and is the dominant OS in mobile devices—that can run on all devices because of the varying hardware configurations, he said. This creates a fragmented space that is more difficult for hardware makers, which have to fine-tune Android for their own devices. At the same time, there can’t be a single Android update that can be sent to all devices.
Contrast that with the PC space, where there is a unified model across hardware, and the same Microsoft Windows OS can run across all devices. It makes development easier and helps drive a more cohesive ecosystem.
“When you don’t have compatibility as being a big deal from a hardware-vendor perspective, you end up having a very fragmented market,” Torvalds said. “We can see what the situation is in ARM today.”
The compatibility issues have made ARM a difficult architecture to work with, he said.
“I’ve been personally pretty disappointed in ARM as a hardware platform, not as an instruction set, though I’ve had my issues there, too,” Torvalds said. “As a hardware platform, it is still not very pleasant to deal with.”
However, it’s getting better, he said, adding that “ARM has gone from being the biggest pain point in the kernel for me to being something I don’t need to care about, and that’s how I want it to be.”
Officials with ARM and its manufacturing partners have been pushing to extend the architecture into other areas beyond the mobile device space, including the internet of things and servers and other data center gear, opening up new areas of competition with Intel. Intel had tried to muscle its x86 chips into the mobile device space, but was unsuccessful and ultimately pulled back on the efforts, opting instead to focus on new connected systems.
A number of chip makers, including Applied Micro and Cavium, have developed ARM-based server chips, but the adoption has been slower than some analysts had expected, and Linaro is a key driver in creating a software ecosystem around the architecture.
ARM soon will have more resources to rely on as it pushes into new markets—tech vendor Softbank has agreed to buy the chip company for $32 billion, one of the latest moves in a fast-changing processor space.