Power Forward with IBM Power Systems in 2018
The increasing focus on AI for business puts the IBM Power Systems platform in a sweet spot in not only 2018 but also in the years ahead.
Image by Jason Griego
By Evelyn Hoover12/01/2017
As pundits make predictions for 2018 tech trends, most expect artificial intelligence (AI) to play an increasing role in business. In fact, a Forbes writer put AI at No. 6 on his list of the top 10 tech trends for the new year. “So much power remains in AI—in everything from customer service and robotics to analytics and marketing. Companies will continue to use AI to surprise, connect and communicate with their customers in ways they may not even appreciate or realize,” the writer said (bit.ly/2yq9yNn).
The increasing focus on AI for business puts the IBM Power Systems* platform in a sweet spot in not only 2018 but also in the years ahead. IBM has invested deeply in both hardware and software for AI and deep learning, especially as the two relate to cognitive computing, according to Stefanie Chiras, vice president, IBM Cognitive Systems Offering Management.
IBM Systems Magazine, Power Systems edition, sat down with Chiras to learn more about what’s on the horizon in 2018 for the Power Systems platform.
IBM Systems Magazine (ISM): Various definitions of cognitive exist. What’s IBM’s definition of cognitive workloads and cognitive systems?
Stefanie Chiras (SC): Cognitive workloads don’t just transact on data; they extract insight from vast amounts of all types of data—private, public, structured, unstructured, streaming, the list goes on. To implement these workloads requires a combination of all types of data sources combined with advanced analytics, and the software paradigm of neural networks used in machine learning and deep learning techniques. These aggressive workloads drive requirements on the infrastructure and commodity hardware just can’t keep up. This underpins the need for cognitive systems, which are designed for high data throughput and extreme compute to deliver workload insights fast enough to take meaningful action.
ISM: How does cognitive relate to AI?
SC: Artificial intelligence—or augmented intelligence as it increasingly being referred to—is really equivalent to the insights from cognitive workloads. At the highest level, it’s about how computing can mimic the incredible flexibility of the human brain to process and create context around massive amounts of data.
While, some stigmas about AI date back to sci-fi books and films, this isn’t about robots and replacing humans. It’s about providing insights from lots of data to better inform businesses and humans to make decisions and take action. Each of us has a certain amount of experience—a certain set of interactions—that we base our decisions and our views upon. Cognitive insights and cognitive workloads provide an additional set of insights pulled from a vast amount of data and experience.
For example if a person suspects he has melanoma, he visits the dermatologist. That doctor will assess, based upon a visual assessment of a mole, its danger and the ramifications. Then the doctor provides a diagnosis. With cognitive capabilities, a photo can be taken of the mole. That photo is analyzed based upon data and models that are created from years of worldwide data—not just what that individual doctor has seen before or been trained to identify. It pulls out insights to make recommendations to the doctor on whether the mole is melanoma.
But it still comes down to the doctor, knowing that particular patient’s history and knowing how they have worked with this patient in the past to make a final recommendation. But now that recommendation is augmented by an additional set of information, based upon data that the doctor normally wouldn’t have access to.
ISM: Why is cognitive gaining traction with IBM clients now?
SC: IBM clients are now realizing the competitive advantages that can come from cognitive insights. The key convergence enabling that transformation is the growing amount of data and the software techniques of neural networks combined with new infrastructure technologies, like POWER8* and now POWER9*, that can deeply exploit acceleration. This combination opens up a whole new world of possibilities.
Growing amounts of data such as sentiment data from social media and inputs from the Internet of Things can augment a business’ core data. And now infrastructure compute capabilities can be supplemented by acceleration and software capabilities that come together to take all of that data and pull out insight.
The combination of the changes in types and amounts of data combined with changes in technology—both hardware and software—allows clients to transition from the age of transactions to the age of insight. It’s about pulling insights—rapid, accurate insights—out of lots and lots of data.
ISM: What’s the role of the Power Systems platform? Does infrastructure matter?
SC: In the ability to deliver on cognitive workloads, the infrastructure matters—it absolutely matters. Infrastructure designed for data and acceleration has enabled the revolution of cognitive that we see today. In addition, however, it has to be easily consumable and aligned to the software and the tools that developers use.
IBM has built the POWER* architecture to be a beast for data. It consumes data; it processes data; it drives data throughput; it just plain works on data better than any other architecture in the industry. We’ve proven that capability in our AIX* and IBM i offerings for years and now we’ve taken on the demands of Linux* and cognitive workloads.
The POWER8* processor represented a huge step, taking data throughput to a new level and enabling broader participation in the Linux ecosystem with little endian support. It also took system-level computing a leap further by talking to accelerators in a differentiated way. Through interfaces like Coherent Accelerated Processor Interface (CAPI) and NVLink, acceleration technologies like field-programmable gate arrays (FPGAs) and graphical processor units (GPUs) can be exploited with POWER in an industry- leading way.
The POWER9* processor takes that strategy to the next level. With POWER9, IBM will deliver a leap in data throughput and compute capability, which will bring huge value to AIX, IBM i and Linux clients. The acceleration capabilities vault forward as well. CAPI and OpenCAPI will have greater bandwidth, and NVLink for GPUs will double its bandwidth capabilities.
“IBM’s goal is to bring cognitive capabilities to every client. It will be imperative for clients of all sizes to participate in this age of insight if they want to differentiate themselves”—Stefanie Chiras, vice president, IBM Cognitive Systems Offering Management
But IBM Cognitive Systems isn’t just about delivering the hardware in isolation. The portfolio delivers value at the full solution level as well. PowerAI takes our most advanced GPU-enabled hardware and delivers with it the leading open frameworks that deep learning models are being built with today. IBM-supported tools like Tensorflow, Caffe, Theano, Torch and Chainer, which are optimized for the accelerated POWER platform, are available for download. IBM Research has delivered further software advances such as Distributed Deep Learning, which leads the industry in the ability to scale GPU workloads across a cluster.
ISM: What role does OpenPOWER play in cognitive?
SC: The launch of the OpenPOWER Foundation at the end of 2013 continues to be key to IBM’s strategy.
IBM embraced the view of open and how to build a system-level platform that’s the best for cognitive workloads. That allows us to do what we do best on the processor and allows others like NVIDIA to do what they do best in the GPU technology. As a result, the client gets unprecedented system-level value. We’ve created technologies like NVLink, which allows the transfer of data between the CPU and GPU to be up to 5X faster.
Because we’ve taken a systems-level approach for machine learning and deep learning techniques, clients can leverage the best innovation in acceleration and CPUs. The joint development of technologies enables them to work better together. That’s further augmented with the PowerAI software optimizations.
That’s not done anywhere else in the industry and it’s really foundational to what we believe is the value of being open. It’s about having a community of innovators developing upon the architecture. IBM’s work for the U.S. Department of Energy to deliver supercomputers to Oak Ridge National Labs and Lawrence Livermore National Labs leverages unique technology that came as result of OpenPOWER partners NVIDIA, Mellanox and IBM jointly innovating, jointly delivering value.
As clients participate in the broader ecosystem of cognitive workloads and create models around their machine learning and deep learning, they get the best technology and simple access to ecosystem-based tools and capabilities.
ISM: Is cognitive important to smaller clients?
SC: For every company of every size, cognitive is going to become important. IBM’s goal is to bring cognitive capabilities to every client. It will be imperative for clients of all sizes to participate in this age of insight if they want to differentiate themselves.
IBM Watson* solutions offer a great opportunity to access pretrained cognitive models as a service. I’ve spoken with many clients who have taken what they do transactionally on their own premises and observed an unexplained anomaly in the behavior in their data. These clients have augmented their analytics with Watson capabilities as a service and gained deeper insights. We’re seeing clients realize “the art of the possible.”
With IBM Cognitive Systems, any client can develop their own machine learning and deep learning models on premises using leadership capabilities like distributed deep learning and open industry frameworks in PowerAI. They can also leverage the capability from our partners like Nimbix, which has taken our best servers for accelerating cognitive computing and preloaded them with industry leading open-source cognitive frameworks.
More and more clients are starting to test the waters with what they can do with cognitive. The value of our portfolio is not just about the technology that’s in it but it’s also about the support capabilities that we offer. If clients take their cognitive capabilities into production, they want to have clear line of sight that they can do that with the best innovation and with the best support. Leveraging solutions like PowerAI offers them a chance to experiment with cognitive while knowing that a road map, backed by the full support of IBM and the innovation of the full OpenPOWER community, exists to bring that into production.
ISM: Besides cognitive computing, what else can readers expect from Power Systems in 2018?
SC: IBM will be delivering POWER9 technologies, both accelerated and nonaccelerated, to the Department of Energy in 2017. We will also deliver capabilities built upon POWER9 to market at the end of 2017. In 2018 the POWER9 processor will be brought to the entre portfolio.
IBM remains fiercely committed to our clients on AIX and IBM i. Both OSes are vital for IBM Cognitive Systems and provide a key data source clients will use in their cognitive workloads. We made some exciting announcements in October 2017 around cloud 2.0 for IBM i and AIX, bringing AIX to the IBM Cloud. We will continue to drive innovations into AIX and IBM i with POWER9 in 2018.
Evelyn Hoover is the content director of IBM Systems magazine. She can be reached at email@example.com.More →
Post a Comment
Note: Comments are moderated and will not appear until approvedcomments powered by Disqus