IBM Forms OpenPower Consortium, Breathes New Life Into Power
August 12, 2013 Timothy Prickett Morgan
Back in July, I started what will eventually be a series of stories on what IBM should be doing with its systems business. And like a bolt from the blue, the company did something that I have been mulling for the past several months in earnest and noodling for the past several years off and on. And what IBM has done is to mimic the ARM collective and open up the intellectual property surrounding its Power chips to help foster a broader ecosystem of users and system makers.
I know what you are thinking. Isn’t it a little too little and a little too late? Shouldn’t IBM have done this a long time ago? Like maybe a decade ago? That was when Apple was still using PowerPC chips in its Macs and the three big game console makers–Sony, Nintendo, and Microsoft–were all building machines based on the Power architecture. Big Blue was building a hybrid supercomputer for the U.S. Department of Energy based on a combination of Opteron processors from Intel and Advanced Micro Devices and Cell PowerPC chips, the BlueGene family of parallel supers was doing better than many expected, and the embedded PowerPC biz was doing relatively well, too, with chips in all kinds of devices and places and largely usurping the position of the venerable Motorola 68000 series of chips.
As my wife’s late Uncle Harry Cornell, who was a CFO and financial planner, was fond of saying: “The best time to do anything was 10 years ago. The second best time is right now.”
IBM seems to have finally gotten that message about its Power platform, and is doing something about it by launching the OpenPower Consortium. The organization is a bit skinny at the moment, but IBM is hoping to build an ecosystem of customers who want to license Power cores and other aspects of the machines that Big Blue has created to create their own heavily customized solutions. What IBM is not doing is letting go of the Power instruction set, any more than ARM Holdings has done with the ARM architecture that has taken smartphones and tablets by storm and which is making serious inroads into the embedded market, giving both Power and MIPS chip suppliers more headaches than they are used to in this niche space.
“Clearly, what everyone wants is for us to maintain architectural control,” explains Brad McCredie, who is chief technology officer for IBM’s Systems and Technology Group. “That is a very important piece of this.”
McCredie, who has managed the designs of several previous Power processors, says that in the past, companies wanting to clone a chip architecture have wanted to add instructions or take things out to streamline their designs. This caused contention between partners. Like, for example, the partners in the PowerPC alliance: IBM, Motorola, and Apple. Or even Intel and Advanced Micro Devices, who want to differentiate from each other with their X86 designs but who also need to maintain compatibility. These days, says McCredie, the market has shifted and customers–meaning traditional chip licensees as well as big customers who can afford to indulge in custom hardware–want to innovate around processor cores and putting accelerators, peripheral buses, and interconnects on the same die as the cores.
And so, with the OpenPower Consortium, IBM is opening up the intellectual property around its Power chips and slapping a For Sale sign on it, with the hope that it can not only make money peddling this intellectual property, but foster a slew of derivative Power designs that may eventually be pumped out of IBM’s own chip plant in East Fishkill, New York. With the game console business winding down, IBM really needs to generate extra chip fabbing business to justify its investments in the 22 nanometer technology that will be used to make the Power8 chips next year and other equipment it will need to push transistor sizes down even further.
McCredie tells me that the OpenPower Consortium is starting out with the licensing of technology relating to the Power8 processor, but that if anyone wanted to license earlier Power designs, IBM is fine with that.
And, IBM is also willing to help customers who want to create their own variants of Power processors figure out how to make those designs work at other chip fabs. There are not many big ones left, but those operated by Taiwan Semiconductor Manufacturing Corp and AMD-spinout Global Foundries are the two obvious ones. Intel has made it pretty clear that even if it does open up its foundry to outsiders, it will not enable direct processor competitors with its wafer bakers.
At the moment, the OpenPower Consortium has four members outside of IBM. They include hyperscale data center operator Google, graphics chip maker Nvidia, networking and switch chip maker Mellanox Technologies, and motherboard maker Tyan.
Google already gets semi-custom chips from Intel and probably AMD for its servers and storage arrays, which involves chips rated for different clock speeds and thermal limits for sure and could also include special instructions on the chips that help accelerate its workloads. Intel offers that for a fee or to ensure it gets the server chip business, and only customers who pay (one way or the other) get access to those aspects of the silicon. This golden screwdriver feature upgrade approach will be very familiar to OS/400 and IBM i shops, who remember all too well the 5250 software tax that IBM used to impose on them.
Google was, as it always was, pretty vague about its plans for Power technology, and it seems to me that what Google really wants is to keep its options open and the heat on AMD and Intel to keep innovating and give it the terms it wants on its chips, chipsets, motherboards, and any other components it uses in its systems. Here’s what a Google spokesperson said about the consortium, and this is all that Google did and will say about it:
“We believe in openness and we are looking forward to the innovation that the OpenPower Consortium will bring to the data center hardware and software industry. The consortium has the potential to establish Power architecture as a viable option for applications running within Google’s data centers.”
The spokesperson then added this as an aside: “OpenPower is a consortium of IT companies coming together with the intent [original emphasis] to open the Power architecture to its members in order to bring new innovation to data center hardware and software. It is early days, and we as a member are looking forward to seeing what can come out of this consortium.”
Nvidia is already an ARM processor developer and that is not going to change, the company tells me. So don’t think Nvidia is going to launch a Power processor project. That said, Nvidia and IBM are going to be working together to tightly couple Power processors and Nvidia GPU coprocessors to make very powerful machines for supercomputing and other workloads. I was not able to reach Mellanox or Tyan at press time, but presumably neither company is actually going to license Power cores, either. Mellanox no doubt wants to tightly integrate with the PCI-Express bus on the Power8 chip and Tyan is probably thinking it could make some dough manufacturing motherboards for alternative Power system providers should they emerge.
The important thing as far as IBM i shops are concerned is the mantra I have been chanting for the past several years: Anything that makes the Power chip stronger lets IBM i live longer. So here is to hoping that the desire to have an alternative to Intel (which has pretty much vanquished AMD in the server racket except at big clouds and big banks) is strong enough to let Power in next to ARM at this state in the game.