The OpenPOWER consortium, which met in San Jose, CA, last month, identified key workloads that have an affinity to the IBM POWER processor’s capabilities. Following its founding in 2013, and more than two years into organizing a worldwide consortium, more than 200 members have joined the OpenPOWER Foundation.
Last year’s Summit had 15 products based on OpenPOWER; this year’s had 59. The growing ecosystem shed light on the deployment use cases for OpenPOWER, which leverages IBM’s multi-core RISC processor for a variety of demanding workloads.
Three things drew the consortium together:
- Leveraging IBM’s POWER chip designs, including eight-threaded processing, for scale-out computing and cloud computing
- Support for highly parallel processing, especially for analytics and HPC
- Ability to license OpenPOWER technology, and to build on it
Use-Cases for OpenPOWER
Customer profiles for OpenPOWER adoption include:
- Hyperscale datacenter providers or cloud service providers that have in-house systems design capabilities. Examples: Google, Rackspace.
- Systems vendors based outside the U.S. that want to leverage POWER for their own server designs. Examples: Tyan, Inspur and NEC.
- HPC sites such as large national laboratories and universities. Examples: Lawrence Livermore National Laboratory, Oak Ridge National Laboratory, LSU (Louisiana State University) and Rice University.
Who’s In OpenPOWER
Key members of the OpenPOWER Foundation include IBM, Google, GPU-maker NVIDIA, Mellanox, and a number of systems vendors in Asia, including many from China. All of them support optimizations around the RISC-based POWER processor architecture and related hardware devices. Speakers said they envision multi-platform deployments, combining x86 and OpenPOWER platforms, in datacenters.
Adoption of POWER technology in China is a special space, given that China is looking for deep involvement in systems design. Speeches at the OpenPOWER Summit showed vendors in China plan to build on POWER for differentiated system solutions, in addition to the x86 servers they build/sell. To expand its global presence, the OpenPOWER Foundation plans to add summits in Europe and Asia.
POWER has an affinity for analytics, high performance computing (HPC) and scientific/engineering workloads. By way of background, POWER is the engine driving IBM’s Watson deep analytics – and POWER-based servers excel at analytics, Big Data workloads, and high performance computing.
In datacenters, systems based on POWER generally are surrounded by large numbers of x86 systems based on Intel Xeon technology. OpenPOWER systems vendors are looking to differentiate their products from the high-volume x86 designs, or to have more control over the designs – developing processors for specific, optimized workloads. At customer sites, the emergence of software defined infrastructure (SDI) will make management of this mixed hardware environment more pragmatic, due to the abstraction of software in the datacenter’s management control plane.
OpenPOWER is not expected to become a high-volume challenger to x86 architecture in datacenters. Rather, it will be a platform for optimized systems using POWER features/functions, including:
- Eight-way multi-threading
- CAPI acceleration. CAPI is an acronym for Coherent Accelerator Processor Interface, which links FPGAs directly to the POWER chip. This is a key feature of OpenPOWER system designs.
- Optimization for Linux and open-source software, in addition to IBM AIX Unix and the IBM I operating system (formerly OS/400).
- Wider adoption of OpenPOWER, especially in highlighted market spaces, such as HPC/supercomputing; and deep analytics and cognitive computing – following the patterns developed by IBM for its Watson deliverables.
The OpenPOWER Foundation is finding new, and expanded, ways to bring POWER-based systems to the marketplace. This is building a new type of ecosystem for IBM’s POWER technology, which is best known for its role in IBM’s Power systems and platforms. A larger ecosystem attracts new channel partners and ISVs, and provides greater volume shipments than if POWER systems came from IBM alone. Having higher volume changes the economics of continuing to build and fab these microprocessors – far different from the decades when IBM was the sole provider.
Perhaps the most fascinating role for OpenPOWER is as a new kind of engine for cloud service providers (CSPs). Clearly, these processors will be deployed as part of a wider x86-based infrastructure, but POWER will take on interesting roles for analytics at those sites. We expect that they will be deployed alongside x86 server engines in CSPs, rather than replacing row upon row of already-installed x86-based scale-out infrastructure.
Bottom Line: OpenPOWER technology will surface in Cloud Service Provider and HPC datacenters – and OpenPOWER licenses will allow new market entrants to leverage POWER architecture to build new server and storage infrastructure.