Pages

Thursday, April 9, 2015

Supercomputing’s Future is Open Systems and that means IBM’s OpenPOWER

By Rich Ptak/Bill Moran



Supercomputing is a fascinating topic. There is an organization maintaining a Top 500[1] supercomputer scorecard listing the world’s top supercomputers. In the latest edition (November 2014) of the top 10 systems listed, four are IBM Power-based systems, three are Cray, two are Intel, and one is Sparc-based. Clearly, Power has already earned a strong position at the top of the heap and has momentum there.

Another fair question might be: “what will OpenPOWER’s position be going forward?” We believe the arrival of OpenPOWER makes it stronger. Power’s current position was established with an architecture proprietary to and controlled by IBM. OpenPOWER’s arrival changes that equation for the better. 

Let’s consider some of the way OpenPOWER is strengthened. Of course, all the usual arguments for an Open architecture apply. OpenPOWER has attracted more than one hundred companies to join the OpenPOWER Foundation[2]. All have complete access to the technology; all can design devices to attach to the technology. They can even make their own Power chips and servers as already announced by several Chinese[3] companies (PowerCore, Zoom Netcom, ChuagHe, etc.). All  member companies, universities, and non-computer vendors e.g. Google, can contribute to evolving the architecture to meet their needs. 

Additionally, the situation in the Supercomputer arena has also changed. Past Power successes were achieved by IBM alone. Now other companies are contributing both their technology and their intellectual capital to make OpenPOWER successful. We have already seen IBM and NVIDIA collaborate to win the DOE project. In addition, other foundation companies, e.g. Mellanox and Micron. have made strong contributions to Supercomputer technology. 

How does this position OpenPOWER against the competition?  As we have seen, the key competitors at the high-end are Intel -based and Cray. Cray’s niche at the high end is not in a volume growth position. Intel is competitive across the entire space. Power’s RISC based architecture is its key advantage versus Intel.  The architecture was designed to allow hardware optimization [4]. Intel’s older, CISC-based X86 architecture carries a lot of baggage as part of its PC heritage.  A second Power advantage, Intel[5] has apparently been locked out of future supercomputers that may be constructed in China. As mentioned, Chinese companies can build Power-based chips and servers, something now no longer possible for Intel.

In summary, OpenPOWER begins with a strong, established position in the Supercomputer space. It can build on this momentum. In combination with the intellectual property of foundation members should easily give future products an advantage over the closed, proprietary systems they must compete with.


[1] See https://en.wikipedia.org/wiki/TOP500  and http://www.top500.org/
[4]The first modern RISC system , the IBM 801,was  developed by John Cocke at IBM. See https://en.wikipedia.org/wiki/Reduced_instruction_set_computing  for a discussion.