IBM, Google, Nvidia and Pals Launch OpenPower
By truly "opening" the Power architecture to Consortium members -- particularly in the sense of licensing or sharing IP and allowing the design and production of customized processors -- IBM is pursuing a path distinctly different from and even oppositional to the proprietary control Intel exerts over its own architectures and the chips it manufactures.
08/19/13 5:00 AM PT
No matter how important or well intentioned, technology industry press releases are seldom entirely free of hyperbole. So in parsing any announcement, it's good to keep an eye on 1) what is being done; 2) who is involved; and 3) how and why the world will be different if they succeed.
In the case of the OpenPower Consortium recently announced by IBM, Google, Mellanox, Nvidia and Tyan, the effort is grounded in IBM's decision to make its Power microprocessor architecture and related technologies fully available to partners and customers.
The "open" part of the initiative entails three points, explained Tom Rosamilla, SVP of IBM's Systems & Technology Group and Integrated Supply Chain.
"First, we are licensing the microprocessor technology to other companies openly -- meaning they get to look at the blueprints for the processor and the software ... and will be able to hire IBM or other companies to manufacture the processors and other related chips," he wrote in a blog post.
"Second, the Consortium will harness the open-collaboration business model [under which] companies that join will share innovations with one another and collaborate on specific technology development projects," Rosamilla continued.
"Lastly, the initiative takes advantage of the open-source Linux operating system ... the [OS] of choice in cloud data centers," he concluded.
What are we to make of all this? If you think the Power architecture is already open, in the sense of IBM delivering Power-based silicon to third parties for a variety of products and strategies, you'd be partly right. The company has long sold Power chips on the open market, along with a variety of support and design services. In fact, IBM founded Power.org in 2004 to facilitate just such efforts.
However, OpenPower goes considerably further by providing Consortium members full access not only to IBM CPUs, but also to the entire gamut of Power-related hardware and software IP. Additionally, Consortium members will be free to choose who they like, including IBM, to manufacture the customized Power chips they develop.
The "open collaboration" business model and plan to leverage open source Linux parts of the plan may sound a bit mushy around the edges, but these are areas where IBM is deeply experienced and involved. The company was the first Tier 1 vendor to support Linux back in 1998, has been a solid open source community member ever since, and has built a sizable business around those efforts. In fact, Rosamilla's blog post noted that nearly half the mainframe capacity IBM ships today is Linux-based.
What about who is involved? The Consortium will definitely be judged by IBM's companions, particularly Google and Nvidia, who lend eye-opening gravitas to the proceedings due to their relative and growing positions in cloud/hyperscale and high performance computing.
Switch-maker Mellanox and motherboard maker Tyan are less well-known but represent the kind of companies we believe are likely to recognize the value the Consortium provides and to become active members.
How and why OpenPower might make the world a different place is a bit more complex. Consider how different enterprise data centers are today from how they looked in 2000, just after the run-up to Y2K and before the rundown of the dot-com bust.
Outside of scattered, rapidly aging legacy systems, architectures that once enjoyed major market positions -- Compaq Alpha, HP PA-RISC and IBM AS400 -- are all gone. MIPS, which once drove SGI's Hollywood-celebrated visualization solutions and Tandem's NonStop systems, is now mainly used in technical computing.
In the enterprise Unix/Linux market, IBM's Power Systems owns the lion's share, HP's Intel Itanium-based solutions are bleeding after battles with Oracle, and Oracle's acquisition of Sun has failed to halt UltraSparc's continuing spiral toward irrelevance.
In point of fact, Intel's x86 Xeon processors dominate volume sales in business computing and HPC -- and generation by generation, they are successfully pushing further into traditional enterprise computing territory.
Intriguingly, nontraditional chips are enlivening business computing discussions. Processors like Nvidia's CUDA GPU are extending the boundaries and use cases for supercomputing and HPC. The open ARM architecture is inspiring developments by new silicon vendors, including Calxeda and Marvell, and innovative system designs like AMD's SeaMicro.
Put simply, IBM appears to be aiming OpenPower directly into the center of these developments. By truly "opening" the Power architecture to Consortium members -- particularly in the sense of licensing or sharing IP and allowing the design and production of customized processors -- the company is pursuing a path distinctly different from and even oppositional to the proprietary control Intel exerts over its own architectures and the chips it manufactures.
The Next Generation
At the same time, OpenPower is likely to impact developments in other data center-focused open chip architectures, especially ARM. Why? Because despite the considerable efforts and innovations of their vendors, ARM-based solutions are in an emergent state. Conventional wisdom suggests that should change significantly as 64-bit ARM chips and systems begin entering the market over the next 12 to 24 months. In fact, ARM proponents believe that event will also entice the sizable number of ISVs and other developers the ARM ecosystem badly needs.
In contrast, IBM's Power architecture is a fully mature, fully proven and widely supported technology, with one of the market's deepest benches of enterprise solution developers. Add in IBM's considerable efforts and experience in Linux, and it would hardly be surprising if the OpenPower Consortium's membership grows quickly. We believe the presence of Google and Nvidia -- neither of them shrinking violets when it comes to data center ambitions -- is a testimony to the strength and potential opportunities of OpenPower.
So is the Consortium likely to succeed? The history of some similar efforts invites caution. Despite being openly licensable for nearly three decades, the MIPS architecture has never expanded beyond a few niche markets. In a strategy that parallels IBM's, Sun Microsystems in 2006 launched the OpenSparc project for licensing its T1 and T2 architectures. Oracle continues to manage the project, but the OpenSPARC Web site currently lists just seven commercial partners.
The Power architecture is in a considerably stronger market position, meaning that OpenPower should find it easier to gain traction than those past efforts. Given continuing contractions in the enterprise Unix market, however, it is also crucial for IBM that OpenPower achieve its aims. In essence, the company is offering Consortium members open access to innovative tools for building the next generation of cloud and hyperscale data centers. How they use those tools and OpenPower remains to be seen -- but overall, IBM appears well-positioned for success.