Qualcomm Enters Server CPU Market

Fresh from the leaked memo that Intel is merging its mobile business into its PC client group, Qualcomm is going the other way and has confirmed that it is entering the ARM server CPU market, an announcement made at its analyst day earlier today.

This is a major trend that less than a month ago I reported from the Linley microprocessor conference. You can go and read the whole thing but the money quote is:ARM has a tiny share. But as I reported last year, that is all set to change. The 64-bit ARM v8 instruction set has opened up new markets and almost all embedded vendors are moving their future investment to ARM. However, the time to design-in, ship and ramp equipment in a conservative market means that the crossover will take 5-10 years, but:

• AppliedMicro shipping X-Gene and sampling X-Gene2
• Cavium plans to sample Thunder in Q4 (their current products are MIPS based)
• Feescale sampling LS1 and plans to sample LS2 this quarter
• LSI/Avago/Intel shipping ARM version of Axxia (although presumably this will be short lived now Intel owns that business)
• AMD sampling Hierofalcon for embedded market
• Broadcom shippping StrataGX and developing Vulcan CPU

Now we can add Qualcomm to this list. Since Qualcomm is pretty much doing the most advanced SoCs on TSMC’s most advanced processes, and given that it has its own ARM processor already, this should put it in a good position. As Qualcomm’s CEO Steve Mollenkopf argued, their ability to quickly adopt next-gen manufacturing processes will give it an edge.

It still remains to be seen if “ARM servers” are really a market. It is supposedly driven by demand from internet giants but until Facebook or Google announce that they are building datacenters at scale using ARM-based CPUs the jury is still out. It is also clear that the companies building ARM-based server CPUs cannot all be successful. I would expect only one or two of these companies to achieve true scale. But there is clearly a value proposition. For some tasks, maximum single thread performance is the most important thing and Intel is untouchable there. But for many tasks, such as servicing hundreds of thousands of simultaneous users on the web, raw performance of a thread is probably less important than aggregate performance of the datacenter against the important metrics of power, cost and physical size. Good performance at 10% of the cost, 10% of the power and 10% of the physical volume sounds pretty compelling. The total cost of ownership of a datacenter includes a high electricity bill to deliver power to the servers and another high electricity bill to power the air-conditioning to get the heat out. Reducing that may be even more important than the cost of the server chips.

But Facebook is at least talking the talk. Jay Parikh, Facebook’s vice president of infrastructure engineering said:Qualcomm-based ARM servers gives us the ability to rethink the way that we have built certain parts of our infrastructure.

So Intel is trying to get into Qualcomm’s primary business and now Qualcomm is trying to get into Intel’s.

Wall Street Journal article is here

See also Will ARM Rule the World?

Qualcomm also announced the latest version of its Gobi LTE modem series, the 5th generation named 9×45. It will be out next year. I’m guessing it is build in TSMC 20nm process. It supports speeds of up to 450Mbps which is pretty amazing (as a comparison, when commercial Ethernet was first introduced it ran at just 10Mbps). This requires carrier aggregation, which means that the mobile device communicates simultaneously using several channels. This has the advantage of retaining backwards compatibility for devices that can only use a single channel. It is supposedly lower power than the previous 300Mbps version, the 9×35 and requires less board space (presumably it is a smaller die and a smaller package).

See also Gobi, the Jewel in Qualcomm’s Crown

More articles by Paul McLellan…