If you're Goliath, what do you do about David? This question has likely preyed on Intel over the past few years as it has watchedgain ever-greater success in the processor market.
One way for a tech titan to deal with a pesky start-up is to spread FUD (Fear, Uncertainty and Doubt) around the challenger's technology. Another approach would be for the incumbent to invest some of its ample resources in dominating fringe markets into which the smaller company would like to expand, shutting off its sources of future revenue.
Intel (2011 revenues: $54bn),of its latest low-power 32nm Atom processors, the 'Centerton' S1200 series. These chips will go up against ARM (2011 revenues: $785m) in dense, low-power 'microservers' in modern datacentres.
Closing power gap with ARM
The 64-bit S1200 series range in power consumption from a thrifty 6.1W up to 8.5W, have clock rates of 1.6GHz to 2GHz, and have PCIe 2.0, along with support for enterprise features like ECC and virtualised workloads.
On paper, these chips come within a hair's breadth of current ARM Cortex A9 datacentre chips, such as those used by thermal design point) of 5W.. Calxeda's server nodes have a TDP (
Calxeda's figures incorporate the power footprint of memory and the networking interface card (NIC), along with the processor, so the real power consumption of the processor itself is a tad lower. Intel, meanwhile, is reporting TDP for just its CPUs — no NIC and RAM included.
This illustrates how keen the company is to put out numbers that indicate its chips have a headline TDP figure that's competitive with ARM.
Intel was not able to provide directly comparable figures showing the TDP of the S1200s with networking and memory factored in.
First low-power 64-bit chips to market
However, Intel will be first to market with 64-bit-capable chips, asare not expected to become available until mid-2013. Applied Micro should deliver the first 64-bit ARM chip in mid-2013, while Calxeda expects to get out in early 2014.
Plenty of Intel OEMs announced plans for Centerton servers on Tuesday. A full list was not available at the time of writing, but promotional Intel materials seen by ZDNet included companies such as Quanta, HP, Huawei, Dell, Wiwynn and Supermicro among the 20 'design wins' the new Atom chips have racked up.
In 2011 Intel predicted microservers could make up to 10 percent of the server market, and the company is sticking by this number.
The leading proponent of microserver technology to date has been SeaMicro, a start-up server maker that received heavy promotional support from Intel until it was bought by x86 rival AMD. Other companies have dipped their toe in the microserver waters, including HP.
But what does this mean for MY datacentre?
So, if you are contemplating a microserver based around many low-power processors, should you go with ARM or Intel?
Even though Intel's chips consume a bit more power than ARM's, they have the advantage of being based on the well-supported x86 architecture
This means that enterprise applications will work as expected and no code porting will be required.
On the other hand, if companies have built much of their software infrastructure around the LAMP (Linux, Apache HTTP Server, MySQL, PHP) stack, then little code porting will be required for their applications, which could run well on ARM-based microservers.
One benefit of using ARM chips is that if you have a massively distributed software system, such as one using the Hadoop file system and MapReduce information cruncher, then the nature of the platform means that it could cost less to run it on many low-power ARM chips than on a few power-hungry Xeon processors.
However, Intel has been in the datacentre business for decades, whereas ARM licensees are only just beginning to make a mark. Therefore it's likely that many enterprises will opt for Intel microservers first due to software support and vendor confidence.
Next year Intel will bring out another Atom chip — codename '' — on its 22nm 'tri-gate' chip fabrication process, which should help cut power consumption further, and in 2014 it hopes to make a 14nm chip.
ARM licensees, meanwhile, will probably use existing 32nm processes from chip fabrication specialists GlobalFoundries and TSMC for next year until those companies' 22nm methods mature. For this reason it's likely that, although it may have a power edge now,.