IBM’s OpenPOWER organization has clearly stepped up its game this week with a massive move towards making deep learning and AI efforts far more affordable. The latest announcement was to expand both its Open Source efforts to include TensorFlow—a Google-developed numerical platform designed for AI and deep learning—and significant enhancements to its NVIDIA-enhanced POWER8 platform—the S822LC (as these things get smarter I’m starting to wonder when we’ll stop using letters and numbers for names and just call them “Bruce”). You can read the announcement here yourself.
Let’s chat a bit about what it means.
Building to a Customer Requirement
The early AI and deep learning systems were largely science experiments. Very complex, very expensive, and you not only had to write the software yourself, but efforts weren’t even shared that well between schools—let alone companies—which made them incredibly expensive and held back these systems’ ability to provide timely and cost-effective benefits.
In short what the customers wanted was something that leveraged what others had accomplished, was far cheaper to purchase and provision, and that could be brought into service timely and effectively. Oh, and since training had proven to be a massive problem, something that helped reduce the training time for the new systems.
IBM Stepped Up
With this solution, IBM partnered with a variety of software and hardware vendors to create a unique deep learning AI platform that is far less expensive, with strong performance, and heavily based on open source technology, which would facilitate cross organizational, and cross company collaboration so learning and advancement could become more of a shared experience and far less aggravating and expensive than the siloed efforts which had preceded it.
This now becomes another showcase offering from IBM, custom-tuned and configured for deep learning and AI efforts and utilizing technology acquired from one of the other AI leaders, NVIDIA. For while IBM initially defined this market with Watson, NVIDIA has been working for some time on improving the overall performance of efforts that use its GPU-based strategy to affordably and significantly improve both the overall performance of the resulting solution and significantly reduce the costs associated with bringing it to market.
The TensorFlow part of the effort was inspired because TensorFlow is one of the most well regarded offerings to ever come out of Google and it is relatively independent of the server OEMs. TensorFlow is currently being used successfully on projects ranging from computer vision, to speech recognition, to text analytics.
TensorFlow isn’t the only framework, though. IBM is also supporting Chainer, CAFFE, Theano, Torch, NVIDIA Digits and others. Given this is IBM, this includes related services and downloadable resources and IBM is initially targeting banking, the automotive industry, and retail segments.
The idea of a more affordably priced, open source, well supported solution should be significant. Companies using the solution should be able to better define and develop related solutions while staying inside their more reasonable budgets.
In the end this should result in an increase in both speed of advancement in the segment and the number and quality of related deployments. This is especially timely in the US where the aggressive President Trump effort to move jobs back to the US from overseas will require massive improvements in efficiency to offset vastly different labor costs between regions. These AI and deep learning systems could be the critical hedge that allows firms to comply with the on-shoring of jobs while still remaining cost competitive.
If it takes a village to raise a child, it certainly takes a group of collaborating partners to create an affordable AI/deep learning platform. With this new effort, IBM has stepped up to that challenge and pulled a number of hardware and software partners together to create something special. A next-generation, relatively affordable platform which could well define the near-term future of related volume, high value efforts.
In short, IBM kicked some butt today.