5 Comments

Summary:

Nine months after IBM opened up its Power architecture, it is launching a new line of servers based on the technology and showing off a server built by Tyan that uses the processor too.

After opening up its Power PC architecture last August, IBM on Wednesday unveiled a new line of servers built to challenge Intel’s x86 architecture inside the data center. IBM said it will deploy those servers in its SoftLayer cloud.

IBM and other members of the OpenPower Foundation, which will manage the Power architecture, also showed off a white box server built by Tyan that was running firmware and an operating system developed by IBM, Google, and Canonical. Taken together, these and other announcements made today around the OpenPower Foundation are an effort to keep the Power architecture relevant and to provide a piece of silicon that IBM is hoping can handle big data — the next generation workload everyone’s salivating over in the infrastructure world.

ibmpower

With that in mind, IBM’s new line of Power Systems servers were built on an open instruction set — much like what Facebook has built with its Open Compute efforts for servers and storage, but at the chip level. It’s a similar strategy to what ARM has done by making its IP cores available for license, only IBM is even giving up the licensing revenue. This means any chipmaker could build a processor using the Power instruction set without paying for the underlying IP.

The new silicon, called the Power8 processor, is a massive 12-core processor that has an open interconnect so it can run with other processors, such as graphics processors (Nvidia is member of the OpenPower Foundation) and FPGAs (field programmable gate arrays). This concession to a heterogeneous computing environment is a significant one, especially as large webscale providers such as Google, Facebook and Microsoft look to customize their hardware for their data center workloads.

The fact that Google is participating in the OpenPower Foundation is a significant publicity coup for IBM. Should Google actually use the Power architecture in its data centers, that publicity coup could turn into a business win, as Google is one of Intel’s largest customers. (We’ll make sure we check in with IBM at our Structure conference held June 18 and 19 in San Francisco.)

This isn’t some newfangled chip design. The Power architecture has been around since 1990, appearing in servers, Apple’s computers until it made the switch to Intel in 2005 and other machines. However, the question before IBM and its partners in the chip industry is whether it can become a viable server for the market.

People trying to build on the Power8 architecture will need chip design expertise that goes beyond a CPU, such as incorporating networking on-chip and between chips, other components found on a motherboard and then manufacture them. Any company building a box would also need to build firmware and a link between the Power architecture and today’s data center software.

The Tyan box is a good start to this effort, and IBM’s own servers are using the technology. IBM said the new IBM Power Systems are capable of analyzing data 50 times faster than the latest x86-based systems. However, benchmarks are the most fungible of stats, and I’m sure Intel could find a use case and workload to dispute IBM’s data.

But this new architecture and business model around Power is not just about trying to build a chip that processes data faster. Other companies from Tilera to Intel are all trying to market chips and architectures that can analyze data faster or within a lower power envelope. Demand for workloads including everything from storing data in Hadoop clusters to analyzing it has infrastructure vendors all rushing products to market.

But in opening up the Power architecture IBM’s not just trying to get in on a hot workload, it’s trying to save a more than twenty-year-old processor design from obsolescence. And according to IBM, it has spent $2.4 billion in the last three years to do this. As Patrick Moorhead, from the analyst firm Moor Insights and Strategy wrote in a paper:

“However, IBM historically has been unable to translate their supposed speeds-and-feeds advantage into additional market share in the vital Linux internet-scale market. IBM now is pinning its hopes on adopting an ARM-like open development model with partners to foster more innovation, competition and adoption. IBM hopes to save the POWER processor from the fate suffered by SPARC, Itanium, DEC Alpha, MIPS and other proprietary architectures.”

The question is whether opening up Power and signing up some powerful friends can propel the architecture into the mainstream for data center customers that need to crunch data. Today’s news is a great start, but it’s the follow on from IBM’s customers and its partners that will determine the Power architecture fate.

Updated: The story was corrected at 11:28 to adjust the quote from Moor Insights. The firm said “IBM has been unable to translate their supposed speeds and feeds advantage.”

  1. Um you blew it. The quote was that IBM historically has been UN-able to …, not “able to”!!

    Share
  2. Um Stacey? The quote was that IBM historically has been UN-able to …, not “able to”!!

    Share
    1. Karl, thanks for bringing this to my attention. It’s changed.

      Share
  3. IBM needs to move the POWER architecture to optical processing, and keep the industry
    on its toes. 64 core systems with 4096 bit data path access to optical arrays would make
    the optical computing industry take notice. Once optical processing can interact with the
    gamer/homeboy desktop market, prices not only will drop, but reasonable configurations
    for a.i. and HPC software should be forthcoming. GPGPU’s should be codeable in optical
    arrays..and the POWER architecture’s advantages in SIMD SMMP admissible. True 256-bit
    and higher (1024-bit data addressing capabilities are already in Mac OSX) could finally
    reach into the current bottlenecks and relieve some intractable conundrums. WW

    Share
    1. Shimizu Randall Sunday, April 27, 2014

      This makes a lot of sense since IBM is the current leader in optical chip interconnects.

      Share

Comments have been disabled for this post