While they debuted few whiz-bang offerings, IBM did roll out a handful of products that allow users to run AI technologies across more platforms, including Power Systems servers and competitor clouds.
IBM embellished and simplified its AI and hybrid cloud strategies last week, giving users more machine learning capabilities and integrating Watson technologies more tightly with Power Systems servers.
At the annual IBM Think 2019 conference, the company debuted an improved version of the Watson Machine Learning (WML) Accelerator, formerly known as PowerAI Enterprise, which better integrates with Watson Studio and Watson OpenScale monitoring software.
The WML Accelerator allows users to train and deploy learning models built using Watson Studio and Watson OpenScale faster and those models can be run across multiple cloud environments, according to Sumit Gupta, vice president of AI with IBM's Cognitive Systems group.
The Watson Machine Learning Accelerator marks the first time IBM has built an integrated AI offering that spans IBM Watson and IBM Power Systems, according to Gupta. He believes it is an important step in the company's goal of making AI available anywhere. For end users to use AI anywhere, they need hardware and software that's optimized for AI workloads.
IBM adds Power to Watson
One analyst believes IBM made the right move in tying AI software more tightly with Power Systems hardware, given the compute intensive tasks associated with AI and machine learning workloads.
If you look at what's happening with AI and machine learning now, the need for speed is ramping quickly.
Judith Hurwitzpresident and CEO, Hurwitz & Associates
"If you look at what's happening with AI and machine learning now, the need for speed is ramping quickly," said Judith Hurwitz, president and CEO of Hurwitz & Associates, technology consultants in Needham, Mass. "You can't keep adding massive servers to keep up with performance demands as you add more and more data. So building in accelerators is really important."
The Machine Learning Accelerator, which improves the performance of workloads up to 46 times, according to IBM, also serves as an orchestrator among teams of programmers working on the same or different projects, Gupta said. He attributes the improved speed to the new high-speed interconnects between the Power Systems' CPU and GPU that allows data to be shipped back and forth more quickly.
"If you had 100 data scientists all banging on the same set of resources, some would experience deteriorating performance or get locked out of some resources," Gupta said. "But [the WML Accelerator] automatically unlocks resources not being used by some and transfers them to others who need to complete a task."
IBM Watson extends to AWS, Microsoft
IBM has an opportunity to get a leg up on competitors in the AI market by allowing its Watson AI technology to operate in competitors' environments -- including those of AWS and Microsoft, according to Dana Gardner, president and principal analyst at Interarbor Solutions LLC, based in Gilford, N.H. It gives corporate IT users the choice of mixing and matching AI and cloud technologies from multiple vendors.
"IBM all of a sudden has an advantage here with a best-of-breed AI engine and allowing it to move around public, private and hybrid clouds," Gardner said. "If they are successful in doing this it could give their AI brand more prominence."
Gardner believes IBM's AI anywhere initiative could make the jobs of data center professionals easier by relieving them of having to make multiple and separate decisions about AI and hybrid cloud. It could free them up to focus more on other pressing issues such as data sovereignty, governance and management.
"You don't have to run this all on IBM software or just on a private cloud," Gupta said. "You can run it on AWS or in a hybrid or multi-cloud environments made up of clouds from multiple vendors."
IBM also renamed PowerAI Enterprise to Watson Machine Learning Accelerator to better fit in with Watson Studio and Watson OpenScale brands, while embedding WML into those two products.
"Our [AI] technology has multiple components but now has a single theme built around AI, high performance and analytics," Gupta said. "For instance, we have features in our offering that can take a TensorFlow job that could run for weeks or months and make it run three times faster."
An important barrier to the broad-based adoption of AI is the lack of skilled IT professionals to implement and support the technology, IBM officials admit. This was backed up in Gartner's 2019 CIO Survey where some 54% of respondents cited a lack of necessary staff skills to successfully implement AI projects and 27% citing the complexity involved with integrating AI with their existing infrastructure as a showstopper.
"Even companies ready for AI can't find enough qualified people, and when they do it costs them a fortune," Hurwitz said. "And when they find these people they are typically commanding enormous salaries, which are out of the reach of many companies."
Dig Deeper on Data center ops, monitoring and management
