IBM launches cognitive computing hardware unit: Enter the Watson, Power 9 stack

Here's the upshot: IBM is looking to speed up training for Watson, neural networks and machine learning. It's going to use its research and hardware knowhow as well as the OpenPower ecosystem to do it.

IBM has formed a cognitive systems unit under its hardware umbrella in a long-term move that aims to speed up Watson training time with integrated systems that'll produce insights almost in real-time.

bob-picciano-2.jpg

Bob Picciano, head of IBM's cognitive systems business

The unit, which launched February 1 under senior vice president Bob Picciano. is charged with using IBM's research, Power architecture and ecosystem via OpenPower to create systems better prepared to navigate unstructured data, machine learning and artificial intelligence.

"The economic value metrics for hardware are changing. We're moving from programmatic, procedural and transactional in nature to valuing computer systems on insight. How do I get insight out of workloads?" said Picciano. "Computing architecture needs to fundamentally change."

Picciano said that IBM started changing its Power processor architecture about 5 years ago. Like most technologies in IBM's portfolio, so-called cognitive systems will touch multiple divisions. There will be ties to Watson, cloud computing, software and services. IBM's systems for cognitive computing will be used in the company's own cloud as well as for customers building their own infrastructure, said Picciano.

Add it up and it's clear that IBM is looking to its hardware unit to boost Watson adoption as well as boost growth. For instance, IBM's systems division is one of its smallest units and hardware sales for 2016 fell 22 percent from the previous year. However, IBM is betting that hardware is fundamentally changing and reckons that cognitive, AI and analytics systems will grow from a $1 billion market to $4.6 billion in 2019.

special feature

How to Implement AI and Machine Learning

The next wave of IT innovation will be powered by artificial intelligence and machine learning. We look at the ways companies can take advantage of it and how to get started.

Read More

Here's a look at the key points from my conversation with Picciano.

Why hardware? In many respects, a hardware unit tied to analytics and artificial intelligence is counterintuitive. After all, won't most of these workloads be handled in the cloud? Picciano said that analytics, cognitive computing and artificial intelligence will be consumed in multiple channels. "One thing I know is that our clients and data scientists are all experimenting and have initiatives," said Picciano. "And they are looking for the best infrastructure on-premises and in the cloud." Memory enhancements, advances in graphics processing units, interconnects and bandwidth all provide building blocks for better cognitive possibilities.

What's the role of OpenPower? Picciano noted that IBM opened up its Power architecture and forged an ecosystem with more than 300 partners because advancing the hardware cause requires partners. Partners like Google, Mellanox and Nvidia all provide innovation for the Power ecosystem, he said. OpenPower is also critical to offering specific high performance computing knowhow for genomic workloads and the like. In addition, there are a lot of industries involved with OpenPower and they're all looking to bring analytics to data. "OpenPower is a great way to bring in other cloud leaders. It's innovation protection," said Picciano.

Why a new architecture? Part of IBM's cognitive play revolves around upending Intel's dominance. Picciano argues that IBM's optimized hardware approach will enable systems to better handle neural networks, deep learning and training systems to use inference. "This is the first inning of a longer game," said Picciano, who noted that a 10x improvement in training cycles is an early goal. "When you go to the edge of computing and IoT we want training cycles to be instantaneous with continuous inference and predictive recommendations."

Here are a few key takeaways:

  • IBM is clearly going for the Watson as integrating stack theme and it's in the company's best interest to speed up training times. For some industries like healthcare Watson's training and knowledge base can scale. Other verticals will be more fluid. More computing horsepower will be needed to boost Watson's time to market.
  • Don't forget Picciano's comments about this move being a long play. Picciano noted that hardware is what the computer is and software is what it does. Power systems for cognitive computing will have a heavy dose of both. That reality is why IBM is supporting multiple machine learning architectures. The ecosystem is critical on the hardware and software fronts.
  • IBM will use the cognitive hardware unit to add to its lineup of strategic imperatives. As a result, IBM will have more to talk about on the systems front during earnings calls.
ibm-segment-revenue-2016.png
  • Power 9 lands sometime in the third quarter and it's the linchpin to all of the cognitive computing hardware efforts. The challenge is IBM is an underdog to Intel's market presence in high performance computing today. IBM has paved the way for more Power 9 adoption via the OpenPower ecosystem. Chipmakers find new ways to go faster
ibm-power9-accelerators.jpg
  • Most enterprises are likely to consume IBM's cognitive hardware systems via the cloud. IBM is targeting hybrid deployments, but if there's a unique value proposition to the Watson-fueled hardware stack the average company will get it via IBM Cloud.

Newsletters

You have been successfully signed up. To sign up for more newsletters or to manage your account, visit the Newsletter Subscription Center.
See All
See All