A new set of GPU-powered AI software announced by IBM today brings automation to many of the tedious, time consuming and complex aspects of AI project on-ramping while reducing deep learning training times, according to Big Blue, from weeks to hours with a new, distributed version of TensorFlow running on clusters.

The new PowerAI software is comprised of four primary parts:

• “AI Vision,” a tool designed for developers with limited knowledge of deep learning to train and deploy deep learning models for computer vision.
• Integration with IBM Spectrum Conductor cluster virtualization software that integrates Apache Spark to ease transforming unstructured and structured data sets to prepare them for deep learning training.
• A distributed computing version of TensorFlow, the open-source machine learning framework built by Google, that can run on a virtualized cluster of GPU-accelerated servers, which IBM said cuts learning training time from weeks to hours.
• “DL Insight,” a new tool that helps data scientists to sharpen the accuracy of deep learning models by monitoring the deep learning training process and automatically adjusting parameters for peak performance.

“We’re adding a set of tools to ease development for data scientists and we’re adding a set of features that accelerate the training time,” IBM’s VP, HPC, AI and Analytic, Sumit Gupta, told EnterpriseTech (HPCwire’s sister publication). “PowerAI makes it much easier for data scientists and developers to use AI to build their applications, rather than having to write complicated code, worry about cluster management, and issues of that kind.”

IBM PowerAI software is “curated, tested, and pre-packaged distribution of the major deep learning frameworks,” including TensorFlow, Caffe, Torch, Theano, Chainer, NVIDIA DIGITS, among others. In making the announcement, IBM called attention to what it said are the performance advantages of GPU-driven AI implementations on the IBM Power Systems S822LC for HPC server, for which PowerAI is optimized.

The server combines IBM POWER processors and NVIDIA GPUs, embedded with a high-speed data interface between the POWER processor and the NVIDIA GPU (NVLink), IBM said. This coupling delivers higher performance in AI training, enabling developers to try new models, parameter settings and data sets at a faster pace, according to IBM.

“IBM PowerAI on Power servers with GPU accelerators provide at least twice the performance of our x86 platform,” said Ari Juntunen, CTO at Elinar Oy Ltd, an electronic content management company. “Everything is faster and easier: adding memory, setting up new servers and so on. As a result, we can get new solutions to market very quickly, protecting our edge over the competition. We think that the combination of IBM Power and PowerAI is the best platform for AI developers in the market today. For AI, speed is everything —nothing else comes close in our opinion.”

IBM also cited the example of Korean Electric Power Research Institution (KEPRI), which wanted to use drones for inspection of high-voltage power lines.

“We needed a deep learning and high speed storage platform that could process and store the vast number of images/videos we receive from the drones,” said KEPRI’s Chan-Wook Lim. “(PowerAI) has met those needs, allowing us to improve our system while also providing a cost reduction for our inspections.”

IBM’s Gupta said KEPRI is typical of the kind of computer vision workloads PowerAI and AI Vision is designed to simplify.

“The time consuming part of it is using a framework like TensorFlow on a 100M images,” he said. “You run into a challenge when you have a 100M images that need to be transformed and prepped to be input into TensorFlow. We automate the data prep and ETL using Spectrum Conductor…, it automatically launches underneath a whole cluster of Spark jobs, each one of them is running and transforming 5 million of those images at a time. From a user perspective, they don’t even know that…it went and launched a whole lot of jobs on a cluster, all they know is that the data is getting transformed.”

“Data scientists and an emerging community of cognitive developers will lead much of the innovation in the cognitive era. Our objective with PowerAI is to make their journey to AI as easy, intuitive and productive as possible,” said Bob Picciano, senior vice president, IBM Cognitive Systems. “Power AI reduces the also reduce frustration of waiting and increase productivity. Power Systems were designed for data and this next era of computing, in great contrast to x86 servers which were designed for the client/server