Calxeda, the Austin, Texas-based chip firm that’s using cell phone chips to build servers, has raised $55 million more from new investors that include Austin Ventures and Vulcan Capital, as well as the firm’s existing investors. The company, which just this month said its servers would be in production by the end of the year, had previously raised $48 million in a uniquely structured round designed to get the company all the way to actual customers.
The idea of using ARM-based chips in the data center has gained ground as customers who own giant data centers have focused on making their servers more efficient. In several use cases, from serving web pages to some big data applications, using many smaller cores is more energy-efficient that using a few massive cores. It’s part of a larger trend toward matching the server hardware to the workload, recognized by Calxeda’s CEO Barry Evans (pictured) all the way back in 2008 when he started working on the company that became Calxeda.
The additional funds will no doubt help the company in what will be a slog against a variety of other vendors who also have ARM licensees and see the opportunity in the server market. For example, Calxeda already has deals with server manufacturers such as HP to produce gear that contains its boards, but Dell so far is hedging its bets with a server that uses an ARM design from Marvell.
Patrick Moorhead, president and principal analyst at Moor Insights & Strategy, notes that software expertise and specialization will determine who succeeds in this market. He says via email:
As in smartphones, the most successful ARM-based server SoC [system on a chip] and platform vendors need to bring software expertise to the game. Some vendors will only bring chip technology to the market which won’t really make a dent in the x86 world. Those like Calxeda who are optimizing for the rack and not just the chip will do well in specialized sections of the new class of exascale data centers.
Calxeda’s technology aims to cram as many as 288 quad-core servers using ARM-based chips in a 4u box with a specially made ASIC that handles the network between that multitude of chips. It’s a similar strategy as the one employed by SeaMicro, a company making a similar box using Intel’s low-power Atom chips, which was purchased by AMD earlier this year. As someone wo’s been anticipating the use of alternative processors in the data center since 2008, I’m excited to see the use of ARM cores and even GPUs gaining ground in production environments. Welcome to the era of heterogeneous computing. Looks like Calxeda’s latest investors recognize that it’s here.