Linus Torvalds praises Arm servers, but claims the economics and ecosystem are missing

The creator of the Linux kernel draws a parallel between the death of Unix and the lack of momentum in Arm servers due to a dearth of developers.

Inside AWS Graviton: Arm servers available in public cloud for the first time Amazon's development of a homegrown Arm CPU, combined with new AMD EPYC powered instances announced this month, threaten Intel's hegemony in cloud computing and enterprise servers.

Linus Torvalds posted some harsh criticism against the prospect of wide adoption of Arm servers in a post in the Real World Technologies forums late last week, following Arm's announcement of Neoverse N1 (formerly Ares), an Internet of Things (IoT)/edge computing platform. The crux of the argument was that developers "will happily pay a bit more for x86 cloud hosting, simply because it matches what you can test on your own local setup, and the errors you get will translate better," even for things which are intended to be cross-platform, such as perl, PHP, Node.js, and other similar scripting languages.

SEE: Top cloud providers 2019: A leader's guide to the major players (Tech Pro Research)

Torvalds also points to a "largely theoretical" power advantage, and the fact that Arm is not commercially viable on servers unless it is possible to "deploy more cheaply on an ARM box," with Arm development hardware accessible to developers.

These comments highlight problems with the Arm ecosystem, and inertia caused by the Intel/AMD x86-64 monoculture that has dominated desktop and server computing since the late 1990s. While it is well and good for Arm to have their own "Developers! Developers! Developers!" moment, understanding these claims in context is important.

Arm competitors are aware of this issue, and working to solve it

The need for accessible developer hardware has always been a known quantity—this is not a groundbreaking claim. Raptor Computing—one of the members of the OpenPOWER Foundation, started by IBM to advance the ecosystem around the competing POWER ISA—has been working diligently to provide affordable desktop ATX boards for developers to counteract this exact problem. Raptor offers the full-fledged Talos II Secure Workstation, as well as the Talos II Lite, and further cost-reduced BlackBird motherboards which utilize POWER9 CPUs.

Gigabyte and Phoenics Electronics are also selling a workstation with the Cavium ThunderX2 Arm CPU, appropriately named the ThunderXStation. The system is built around a 4U tower format, and is available with a 32-core/2.2 GHz SoC, either as a single or dual processor configuration, though this is produced in comparatively limited quantities.

Arm needs a UEFI workalike to achieve mass deployment

Because of the Raspberry Pi, the SBC market is littered with cheap and poorly supported Arm-powered boards that lack mainline kernel support. Compared to any arbitrary x86-64 system that can use a Linux distribution using generic install media, OS images for Arm devices rely on complex device trees to ensure the drivers and hardware configuration is performed properly. This is a manual and intensive process.

This can be seen firsthand with Arm-powered Windows laptops, as developers work on an open source project to make Linux usable on those systems. Similar issues can be seen with the LineageOS Android distribution, and the relatively small number of devices it works on. For Android device manufacturers, Google aims to ease these pains with Treble, which provides a hardware abstraction layer to make it easier to manage OS image development across a product portfolio.

Likewise, the Raspberry Pi family enjoys wide adoption due to the ecosystem around it—including OS images. TechRepublic's Nick Heath notes that another competitor like Hardkernel / ODROID "[has] a good reputation for providing stable operating system images for its boards," though the tendency for device longevity to depend on vendor cooperation is a major stumbling block to commodity Arm devices being practically usable and hackable for developers.

The argument is economics, not technology

In the same discussion thread, Linus praised Arm for "nicely strengthening their memory model, to the point that these days it's actually one of the better ones," and that he "[likes] the direction ARM is going with vector math a lot more than the AVX512 that Intel is pushing," stating "I don't know how well it works in practice, but the whole 'let's try to do something that works for different vector lengths' is laudable. I'm quite tired of the model where Intel introduces yet another incompatible model every few years."

For more on Arm servers in the cloud, check out TechRepublic's coverage of AWS Graviton, the first Arm-powered server developed by Amazon, what Arm servers on AWS mean for your cloud and data center strategy, and why Cavium sees a healty future for Arm servers. Steven J. Vaughan-Nichols also offers analysis at TechRepublic's sister site ZDNet about the viability of Arm for cloud servers.

Also see

monsitj, Getty Images/iStockphoto

By James Sanders

James Sanders is a technology writer for TechRepublic. He covers future technology, including quantum computing, AI, and 5G, as well as cloud, security, open source, mobility, and the impact of globalization on the industry, with a focus on Asia.