Just over five years ago, IBM's Watson supercomputer crushed opponents in the televised quiz show Jeopardy. It was hard to foresee then, but artificial intelligence is now permeating our daily lives.
Since then, IBM has expanded the Watson brand to a cognitive computing package with hardware and software used to diagnose diseases, explore for oil and gas, run scientific computing models, and allow cars to drive autonomously. The company has now announced new AI hardware and software packages.
The original Watson used advanced algorithms and natural language interfaces to find and narrate answers. Then, Watson was one supercomputer, but now AI systems are deployed at a grander scale. Mega data centers run by Facebook, Google, Amazon, and other companies use AI on thousands of servers to recognize images and speech and analyze loads of data.
Watson is just one of IBM's efforts; the company has more initiatives to bring AI to other companies. It's releasing more powerful hardware to make deep-learning systems faster while analyzing data or finding answers to complex questions. IBM is pairing those superfast systems with new software tools.
The new IBM hardware, and software tools called PowerAI, are used to train software to perform AI tasks like image and speech recognition. The more a computer learns, the more accurate the results. Training requires a lot of computing horsepower. The new training hardware is now available.
Ultimately, the hardware and software could be the key cog in making Watson technologies easily available to companies through the cloud or on premises. For now, the company is not talking about whether they will be a part of Watson.
The first set of hardware is the Power8 server with the Nvidia Tesla GPUs, said Sumit Gupta, IBM's vice president of high-performance computing and analytics.
The hardware is the fastest deep-learning system available, Gupta said. The Power8 CPUs and Tesla P100 GPUs are among the fastest chips available, and both are linked via the NVLink interconnect, which outperforms PCI-Express 3.0. Nvidia's GPUs power many deep-learning systems in companies like Google, Facebook, and Baidu.
"Performance is very important as deep learning training jobs run for days," Gupta said. It's also important to speed up key technologies like storage and networking, he said.
IBM is also planning hardware and software for inferencing, which requires lighter processing on the edge or end device. The inferencing engine takes results from a trained model, adds additional data or input, and provides improvised results. Drones, robots, and autonomous cars use inferencing engines for navigation, image recognition, or data analysis.
Inferencing chips are also used in data centers to boost deep learning models. Google has created its own chip called TPU (Tensor Processing Unit), and other companies like KnuEdge, Wave Computing, and GraphCore are creating inferencing chips.
IBM is working on a different model for its inferencing hardware and software, Gupta said. He did not provide any further details.
The software is the glue that puts IBM's AI hardware and software in a cohesive package. IBM has forked a version of the open-source Caffe deep-learning framework to function on its Power hardware. IBM is also supporting other frameworks like TensorFlow, Theano, and OpenBLAS.
The frameworks are sandboxes in which users can create and tweak parameters of a computer model that learns to solve a particular problem. Caffe is widely used for image recognition.