Today, as the final installment of our HPCwire People to Watch focus series, we present our interview with Ken King, general manager of OpenPOWER for the IBM Systems Group. Ken is responsible for building and managing the ecosystem of chip, system, ODM, OEM and software partners partners innovating with IBM on top of the POWER architecture.
Ken joined IBM in 1984. Prior to his current role, he served in a wide range of capacities, including general manager of intellectual property, vice president of research business development, and vice president of patents, software and services intellectual property licensing. He was chairman of IBM’s Open Source Steering Committee for two years, and he currently serves as a board member for the Open Invention Network.
HPCwire: Ken, you’ve been named a person to watch in the past – now you’re coming out of 2018 that included the standing up of Summit and Sierra, the two most powerful systems in the world. Please tell us what you and your team hear from users at the two national labs about the two systems.
Ken King: The users are excited and uniformly positive about what they are seeing in terms of performance and availability of the two systems. The Summit system will shortly become available for access to the public through the INCITE program and we expect to see a very broad set of novel applications from the research community and industry make use of the unique capabilities of AI and HPC operating at great scale. The Sierra system is now completely closed to outside users having graduated to serving the mission of Lawrence Livermore National Lab. The Early Science projects will be the first to really use Summit at Oak Ridge.
HPCwire: Update us on the progress of OpenPOWER.
OpenPOWER continues to be a strong and vibrant ecosystem. OpenPOWER technologies and partners were instrumental in the development and delivery of the Summit and Sierra systems that, as you know, are currently the #1 and #2 fastest, smartest and most powerful systems in the world. We are continuing to build support for developers. OpenPOWER member, Raptor Systems is selling an OpenPOWER Desktop starting just over $1200. In December, we launched our first official, but not last, OpenPOWER Training Center. The foundation now has 340+ members and will host OpenPOWER Summits North America, Europe and China in 2H 2019. At our most recent Summit we had over 500 attendees and over 100,000 live views of our technical sessions and keynotes.
HPCwire: Can you give us a hint as to what’s coming down the line for the Power architecture?
Power will continue to embrace hardware acceleration for HPC and AI. As the data requirements for these applications continue to increase, and as core counts get higher, we will significantly increase memory and I/O capability to maintain system balance. We will help our clients get more from their infrastructure by providing machine learning technology that guides traditional HPC applications to reach solutions more quickly, and we will help HPC system administrators gain efficiency by leveraging modern cloud techniques. We’re very excited about the trajectory of our technology and our business as POWER continues to demonstrate industry leadership capability in both HPC and AI.
HPCwire: As IBM fills out its AI-related product portfolio, how would you summarize the IBM AI offering, and how would you compare it to others in the marketplace?
The IBM AI offerings are immense in scope and capability and the byproduct of the work of thousands of individuals across the POWER/Cognitive Systems, IBM Research and external OpenPOWER community. Specifically, in the context of the HPC community we have been very direct that the notion of HPC and AI as separate domains is misleading: HPC and AI need to be viewed as complementary, converging and integrated in a very broad way. For example, we envision AI as a means to help design, guide, and interpret classic HPC approaches to technical problems. We have a raft of research and development projects built around this. Furthermore, while many of the researchers and practitioners of AI focus solely on the complexity of neural net design and new approaches to inferencing, we think there is huge benefit to provide tools that attend to issues surrounding data acquisition and data architecture which represent about 80% of all data science compute activity. We have been testing novel approaches in this arena for about two years and expect to bring products to market this year. In addition, PowerAI and PowerAI Vision continue to demonstrate industry leadership AI/ML/DL innovation and capabilities. We are growing our client base with these offerings exponentially and have enjoyed many significant wins, such as a Top 5 International Airport. We expect even greater growth in 2019.