ARM and the single-purpose server of the future

Servers don't have to do everything to be useful: devices that do just one thing really well will find a niche, too.

Intel is finally making headway with Atom chips that are both powerful (though far less so than its Core range) and power efficient, like the one powering the 8-inch Dell Venue Pro.

Judging from the way executive vice president for strategy Tom Lantzsch talks, that won't worry ARM — and not just because it dominates the tablet market today: ARM has its sights set on at least part of the server market which phones and tablets are driving.

Lantzsch quotes an Intel survey that he says ARM's own research backs up: for every 600 smartphones that are sold, you need a new server; and for tablets such as the iPad that use cloud services, it's more like 125 tablets per server. And ARM will be relevant there as much because of its business model of enabling many vendors to create CPUs as for its specific technology.

Where ARM fits in for servers, he believes, is for single-purpose tasks such as storing the 3 billion photos that were uploaded onto Facebook over a Halloween weekend last year that might not get looked at for more than a few days but won't be deleted.

"They want to offer you the opportunity to retrieve that photo you took in 2006, they really don't care if it takes you 300ms longer to get it and you probably don't notice," Lantzsch says. "But they do want to make sure the cost to contain that is under control. So they're developing server technology that just shuts itself down until you need to do this retrieval."

Read this

Microservers: How ARM can challenge Intel by befriending Facebook

Can the emerging microserver market give ARM a way to compete with Intel in the server space, with the help of web giants such as Facebook?

Read More

For something like that ARM doesn't need to drive a general-purpose server to be useful. "It's not a generic thing at all, it's just storing pictures."

Think about how many ARM chips are being produced, at prices Lantzsch points out are far below what it takes to design, manufacture and sell a general-purpose server chip.

"We will have seven or eight people doing server-based chips that are optimised for specific examples, and that drives the cost of [developing new] server chips down from $300m to $30m," he says.

"When you can make $30m server chips and you can make money doing that, you can address specific issues for customers that are unique to them. We see our role is having a common core with a big software base that lots of people can leverage off that enables this kind of work."

Low-price single-purpose server that spends a lot of time turned off; that's a completely different approach from the usual server principle of virtualising workloads so you can run multiple things on a single server. It makes most sense for companies that have those specific issues, but as that scale of computing becomes common it's going to be a bigger market.

Read more