Official: Arm-based VMs available on Microsoft Azure

Microsoft has opened up Arm-based virtual machines for general availability on its Azure cloud platform, after previewing the technology earlier this year. The Redmond outfit said that Arm-based virtual machines will be generally available from September 1, initially in 10 Azure regions around the world. Also in preview will …

[Pirate Dave]

Why?

Unless these Arm-based machines are significantly cheaper than x86, what's the point? Is it just so MS can look hip and cool and say they have Arm in their cloud?

Not trying to start a processor/architecture pissing contest, I'm just genuinely interested. If it's a VM, what difference does it make if it's Intel or Arm running it? Are there things that Arm just does better (and that would translate through the layers of virtualization)?

[jglathe]

Re: Why?

Interesting question. For running testing on this ABI, maybe. Maybe customers ask for it. If its cheaper, good. As for running Windows on aarch64... thats one big cf I don't understand. You can run Linux (even GUI) on aarch64 already, for peanuts. I do for fun, and because it is quite power efficient, and you learn one thing or two. Could easily go as daily driver with a big iron in the background. So... maybe a preparation for testing their own migration to other h/w, and monetize it, too.

[Pirate Dave]

Re: Why?

I'm not an Arm guy, hence my original question, but isn't one of Arm's big selling points lower-power use? Which is fine in your own server closet or living room, but, eh, does that benefit carry over very well into Microsoft's datacenter within Azure?

"maybe a preparation for testing their own migration to other h/w"

But isn't that anathema to Azure? I mean, that seems like the exact opposite of what Microsoft hopes to achieve with Azure.

[Anonymous Coward]

Intel has historically achieved extremely high profit margins on its server chips. Intel charged thousands of dollars for each chip.

By introducing Arm server processors from multiple suppliers, it introduces competition into the server market which lowers the price of server chips, crushing Intel's margins, and improving price/performance.

Arm's 64-bit architecture was a clean sheet design announced in 2011 and based on RISC principles. There are technical arguments that it should be possible to implement this architecture with better power per MIPS or better power per FLOP. Power consumption is very important in huge cloud datacenters, because in addition to having to pay for electricity for the servers, there is an additional cost of cooling the data center with air-conditioning or pumping cold water through the data center. This means that power efficiency of server chips makes a big difference to overall cost of running the data center.

These technical advantages might help somewhat with cost. I think the bigger story is the crushing of Intel's historically huge profit margins via competition.

Note that Amazon's AWS cloud provider introduced servers using the Arm architecture a few years ago. AWS uses the brand Graviton for these servers, and they are already at their third generation called Graviton 3. These server chips are designed by AWS themselves under license from Arm, and manufactured at a foundry (probably TSMC). This allows AWS to specify exactly what they want on the server chip to fit in their cloud data centers.

In addition to AWS and Azure, other cloud providers have announced Arm servers, including Oracle Cloud, Alibaba, Cloudflare.

This is a historic turning point, with in server architecture shifting shifting to Arm.

Arm uses the brand Neoverse for their server chip designs.

See: https://www.arm.com/solutions/infrastructure/cloud-computing