IBM PowerAI workstation eases AI app development, lowers cost

IBM has delivered a low-cost workstation tuned to work hand in hand with its Power AC922 server to lower the bar of entry for AI application development.

IBM has delivered a low-cost workstation along with an AI starter kit that runs in concert with the Power Systems AC922 server to make it easier for users to develop and deliver machine learning apps.

With the new IBM PowerAI workstation, the company becomes the latest entrant in the race to push artificial intelligence and machine learning technologies closer to where data and data scientists live.

"The more power you bring to data scientists to experiment with different models and data sets, the better," said Chirag Dekate, a digital strategies and AI infrastructure analyst at Gartner. "It allows data scientists to be closer to the problem and also not get bogged down by infrastructure issues."

The low-cost Power9-based workstation, called the Power AI DevBox, is priced at $3,449. The system is built, which was developed in collaboration with Raptor Computing Systems, for corporate and third-party developers to create AI applications. It contains an in-the-box, 30-day free license for IBM's PowerAI Vision. The latter offering helps developers cobble together new AI models.

"Before buying large servers [for AI-based projects], end users are experimenting first with desk-side systems," Dekate said. "Nvidia's DGX Station is trying to do something similar to IBM's product. They hope to bring extreme compute capabilities enabled by GPUs in a form factor for data scientists to work with."

Other low-cost workstations include Lambda Labs' 2- and 4-GPU Workstation, Bizon's Deep Learning Workstation line and Dell's Precision lineup.

Lowering the price of entry to AI development should add incentive to inexperienced users, believes Sumit Gupta, vice president of AI and machine learning with IBM's Cognitive Systems unit. The sticker shock of many of today's offerings can stop IT buyers in their tracks.

"You walk into user shops and say, 'Here is my AI software, but you need to buy this $60,000 system to get started,' and they say, 'Are you kidding me'?" Gupta said. "This isn't just an IBM problem, it is an industry problem; GPU accelerators and AI in general is expensive."

Accompanying the DevBox is the AI Starter Kit, comprised of GPU-accelerated IBM AC922 Power servers with IBM's Watson Machine Learning Accelerator software pre-installed. Other pre-installed software includes open source AI software frameworks and management software that can be used simultaneously by multiple data scientists and the AI-based training tasks they run.

One of those frameworks is Snap Machine Learning, which can be used for the online retraining of AI models, model selection and parameter tuning, and scaling large data sets for things such as recommendation engines, advertising and credit fraud.

Efforts to simplify AI go beyond products

IBM officials said they are keenly aware of the lack of education on the part of many enterprise IT users. Many are confused about the best use case for their individual needs and the ROI of AI.

This drove IBM to start a series of meet-up programs where IBM executives and users get together and discuss the practical implementations of AI and machine learning and how the technologies are relevant to individual users' needs.

"We have focused a lot on educating the user community in these meetups," Gupta said.

Despite the complexity of AI and machine learning technologies and dearth of knowledge in those areas, IBM is confident its latest offerings will allow many users to take a do-it-yourself approach to building their own customized AI-based applications. Gupta claims the Starter Kit eliminates the messy and time-consuming task of pulling together third-party add-on pieces.

The whole industry is moving in the direction of simplifying AI and machine learning, trying to flatten the learning curve needed to adopt it.
Chirag DekateAnalyst, Gartner

"Users can get confused when they realize they have to download [Google's]TensorFlow, or find out they have some Nvidia development driver dependency, or some third-party open source software," Gupta said. "We have cleaned all that up; the systems come with all that pre-installed."

Some observers believe, however, there won't be that many adventurous enterprise users brave enough to develop and launch their own customized app or service without some outside professional consultation. In fact, from a monetary standpoint, IBM is counting on users needing the assistance of its Global Business Services unit.

"Today it's possible to flail away and pull something together, but it will take longer," said Frank Dzubeck, president of Communications Network Architects, a consultancy in Washington, D.C. "[IBM] won't force services on you but many will come to the realization they'll need assistance to kick-start an AI project by themselves."

Gartner's Dekate is more optimistic, saying major players in the AI and machine learning markets are making their respective technologies easier to use so that customers can figure things out for themselves.

"The whole industry is moving in the direction of simplifying AI and machine learning, trying to flatten the learning curve needed to adopt it," Dekate said. "Many more than you think can figure it out."

IBM officials said they will start taking orders for the AI Starter Kit now, but it won't be ready to ship until later this month.

Dig Deeper on AI infrastructure

Business Analytics
Data Management