Opening Power to partners is a good move by IBM, says Peter Judge. But Google’s role may be crucial to OpenPower
IBM’s announcement of the OpenPower consortium is a bold move designed to halt the decline of the Power architecture while the tech giant readies the next generation, Power 8. Big Blue hopes to get Power-based chips established as a third option for servers, alongside Intel’s x86 architecture and ARM.
For many people, it’s almost a surprise to know that other architectures besides Intel’s x86 instruction set (from Intel and AMD) are established on servers. But IBM’s Power 7+ has been well received, and IBM is a leader in the Unix server business which has long been the natural home of RISC processors such as Power and Oracle’s SPARC. It’s just that Unix servers aren’t expanding the way x86 servers are in data centres. IBM’s Power revenues are declining.
The Power of Love?
What Power needs is partners. The architecture already has users outside of IBM, but its reach is reducing. It was once the processor inside Apple’s Macintosh machines, but Apple replaced it with Intel in 2006. It was also widely used in games consoles, but AMD took over there.
The data centre market is expanding, and big players there want an alternative to x86. IBM’s move would have been perfect a year or two back, but it’s now seen as trailing behind another contender – ARM Holdings, whose low-power architecture is coming in a 64 bit version in 2014, which AMD is turning into server chips. Dell is looking at the idea seriously, and specialists like Boston are actually shipping servers.
IBM’s move clearly takes inspiration from ARM. It will make the Power architecture available for third parties to design chips around. That is a good thing, as right now, it has become possible to cram so much onto a chip that designers are spoilt for choice. IBM’s ideas for Power are fine, but what if another player had a different idea which was just as good? This variety is what makes ARM’s model so strong, and IBM hopes to foster this for Power.
The idea will stand or fall on the partners who choose to get involved, and their level of commitment. And that will depend on them choosing Power over ARM.
There are a couple of factors to influence that choice.
Firstly, the two are travelling in opposite directions. ARM has a massive stronghold in mobile and embedded devices, and is moving into servers because the low power requirement of its designs turn out to be useful there too. IBM’s Power has proven it can do embedded duties, and it is starting out with an existing heritage in servers – it could look like a very good option for this reason.
Secondly, ARM does not have any fabrication plants – it’s a fabless chip designer. IBM has a a powerful chip manufacturing business, rivaling that of Intel in many ways. This is a benefit – any partners can be sure of production of their Power-based designs. However, it might deter bigger partners like AMD who would prefer to be making their own chips.
Nvidia, for instance, has put its name to the Power Consortium, but has a strong relationship with ARM already.
Google, on the other hand, could fit very neatly into the Power mould. It has a huge demand for data centre servers, and is not averse to designing its own technologies, and specifying details of its servers.
It could very well use ARM servers, but would find itself working with a different chip builder (AMD, say) instead of Intel.
In the Power model, Google could actually become a chip designer, and order up silicon from IBM’s factories, giving it a huge amount of something Google likes to have – control. Its public statements on the subject have not given much away, talking about Power’s “potential” for a place in Google’s infrastructure.
At this stage there is no deal to move any of Google to Power-based servers. But Intel will be very nervous about that possibility.
Do you know about IBM? Try our quiz!