Calxeda's failure was its own, not an indictment of ARM on servers

The company made a big tactical mistake. Don't let that cloud your view of ARM in servers.


Last week, Calxeda announced what it called a restructuring and what everyone else sees as an implosion. Basically, the Austin, Texas company is dead. Its investors ran out of patience and now what's left is a fire sale. There should be some IP worth scooping up, most notably its fabric designs. Intel might even want them. But the Calxeda experiment is over.

Caldexa tried to be an Intel of ARM server chips. For the unfamiliar, ARM does not make chips. It makes processor designs, which its 100+ licensee – which includes Apple, Samsung, Qualcomm, Nvidia, Marvell and many more – then take and add their own IP around it. Nvidia added its GPU technology to create the Tegra processor, while Apple did its own thing to come out with a 64-bit version of the processor before anyone else.

ARM is known as a smartphone and tablet processor, but in recent years there has been a push on to put ARM processors in servers. The argument was logical: some server tasks don't need a 75 watt Xeon, they can be adequately performed with a 10 watt ARM (or Atom).

But Calxeda made a huge tactical blunder. It focused on making 32-bit server chips, not 64-bit chips. ARM processors are already 32-bit, but they lack a lot of things servers demand, like ECC code for memory problems, reliability and failover features in case an app or a piece of hardware fails, and so on. So there was work to be done to take the basic 32-bit processor you got with an ARM license and make it work in servers.

The fact is, the 32-bit ship has sailed for servers and no one wants to go back. No one is interested in 4GB memory limitations. With 4GB you have no virtualization and no cloud computing. You are back in 2001 with underutilized servers that have enough memory to load the OS, an app or two, and handle a few connections.

The industry was already clamoring for a 64-bit ARM processor, and ARM does have one in the works for next year. Plus, Apple already has one of its own, for its own use, while Qualcomm is working on one as well. The demise of Calxeda won't slow those projects down. If anything it may accelerate them, now that no one is splintering the market any more.

So once again, Calxeda failed. ARM on servers did not. And Calxeda will likely live on. Its assets will be bought out, possibly by Samsung. It has an ARM for server project going based in... Austin, Texas. How appropriate

ITWorld DealPost: The best in tech deals and discounts.