It's been nearly a year since I published my first Special Report on artificial intelligence and urged readers to buy the processor maker NVIDIA (NVDA) at $68.80. With US annual auto production at 17 million, and global car and commercial vehicle production at a record 94.64 million, that is a lot of processors. All new AI startups comprise small teams of experts from private labs and universities financed by big venture capital firms like Sequoia Capital, Kleiner Perkins, and Andreeson Horowitz. The global artificial intelligence market is expected to grow at an annual rate of 44.3% a year to $23.5 billion by 2025.
We're excited about the launch of NVIDIA's Volta GPU accelerators.Together with the NVIDIA NVLink "information superhighway" at the core of our IBM Power Systems, it provides what we believe to be the closest thing to an unbounded platform for those working in machine learning and deep learning and those dealing with very large data sets. Servers with POWER9 and Volta, with its second-generation NVIDIA NVLink, PCI-Express 4, and Memory Coherence technologies, and unprecedented internal bandwidth, will blow people away. Our IBM and NVIDIA partnership around these new technologies will surface for the first time in the U.S. Department of Energy Summit Supercomputer at the Oak Ridge National Laboratory and the Sierra Supercomputer at the Lawrence Livermore National Laboratory, which are pushing the boundaries of big data science and simulation. AI applications data distributed deep learning gpu hardware IBM Cognitive Systems ibm power systems ibm power9 ibm powerai ibm research Lawrence Livermore National Laboratory Memory Coherence NVIDIA NVIDIA NVLink NVIDIA Volta Oak Ridge National Laboratory PCI-Express 4 power systems power9 Power9 chip powerai Sierra Supercomputer software Summit Supercomputer U.S. Department of Energy Your email address will not be published.Required fields are marked *
And since model training is an iterative task, where a data scientist tweaks hyper-parameters, models, and even the input data, and trains the AI models multiple times, these kinds of long training runs delay time to insight and can limit productivity. The IBM Research team took on this challenge, and through innovative clustering methods has built a "Distributed Deep Learning" (DDL) library that hooks into popular open source machine learning frameworks like TensorFlow, Caffe, Torch and Chainer. Figure 1: Scaling results using Caffe to train a ResNet-50 model using the ImageNet-1K data set on 64 Power Systems servers that have a total of 256 NVIDIA P100 GPU accelerators in them. This release includes the distributed deep learning library and a technology preview for the vision capability that we announced in May.
Calling its new invention "the jet engine of deep learning," the company recently announced its Distributed Deep Learning (DDL) library for PowerAI, which hooks into TensorFlow (an open-source environment originally developed at Google), and other deep-learning frameworks such as Caffe, Torch, and Chainer. I previously wrote about a project involving some of my colleagues using deep learning with IBM Power Systems to examine thousands of radiological images in 0.03 percent of the time it takes radiologists to perform the task. The company cites a program in which it incorporated 64 IBM Power Systems servers and 256 NVIDIA Tesla P100 with NVLink GPU accelerators to train the ResNet-101 visual recognition system on the ImageNet-22K data set, shrinking execution time from 16 days to seven hours. The Industrial Age and the more recent Nuclear Age have shown humankind's gift for creating powerful tools, coupled with a difficulty -- even inability -- to control them.
Today, IBM Research announced a new breakthrough that will only serve to further enhance PowerAI and its other AI offerings--a groundbreaking Distributed Deep Learning (DDL) software, which is one of the biggest announcements I've tracked in this space for the past six months. Most AI servers today are just one single system, not multiple systems combined. To paint a picture, when IBM initially tried to train a model with the ImageNet-22K data set, using a ResNet-101 model, it took 16 days on a single Power "Minsky" server, using four NVIDIA P100 GPU accelerators. To top it all off, IBM says DDL scales efficiently--across up to 256 GPUs, with up to 95% efficiency on the Caffe deep learning framework.
IBM claims it has the fastest cloud for deep learning and artificial intelligence (AI) after publishing benchmark tests which show NVIDIA Tesla P100 GPU accelerators on the IBM Cloud can provide up to 2.8 times more performance than the previous generation in certain cases. The tests, when fleshed out, will enable organisations to quickly create advanced AI applications on the cloud. For the IBM tests, engineers trained a deep learning model for image classification using two NVIDIA P100 cards on Bluemix bare metal, before comparing the same process to two Tesla K80 GPU cards. "As the first major cloud provider to offer the NVIDIA Tesla P100 GPU, IBM Cloud is providing enterprises with accelerated performance so they can quickly and more cost-effectively create sophisticated AI and cognitive experiences for their end users."
IBM (NYSE: IBM) today announced a significant new release of its PowerAI deep learning software distribution on Power Systems that attacks the major challenges facing data scientists and developers by simplifying the development experience with tools and data preparation while also dramatically reducing the time required for AI system training from weeks to hours. The tight integration of IBM POWER processors and NVIDIA GPUs is enabled by the NVIDIA NVLink high-speed interconnect. PowerAI's curated, tested, and pre-packaged distribution of the major deep learning frameworks run on IBM Power System severs built for AI. PowerAI is an enterprise software distribution of popular machine learning and deep learning open-source application frameworks.
The new systems tap the Nvidia NVLink technology to move data five times faster than any competing platform, said Stefanie Chiras, an IBM vice president, in an interview with VentureBeat. Collaboratively developed with a variety of tech companies, the new Power Systems target A.I., deep learning, high performance data analytics, and other compute-heavy workloads, which can help businesses and cloud service providers save money on data center costs. "The open and collaborative model of the OpenPower Foundation has propelled system innovation forward in a major way with the launch of the IBM Power System S822LC for high-performance computing," said Ian Buck, vice president of accelerated computing at Nvidia, in a statement. The two additional LC servers available today -- the IBM Power System S821LC and the IBM Power System S822LC for Big Data -- can also leverage GPU acceleration technology to increase system performance levels on a variety of accelerated applications.
It has introduced for self-driving vehicles what NVIDIA describe as their open AI car computing platform (called NVIDIA DRIVE PX 2). IBM has joined forces with a company called Numenta to develop a memory base algorithm called Hierarchical Temporal Memory (HTM) running on specialised hardware. In other words, they will create hardware with massive interconnectivity, which can change their "plastic" network topology, just like human biological neural networks can create intelligence through their neural plasticity. The point is that within five years, car builders will dispose of the right hardware, software and system architecture to design cognizant driving entities, with self-learning capacities and inferring faculties, which will induce over time more intelligent and much safer driving behaviours than the ones we human can show today.
It pairs IBM's Power8 processors with Nvidia GPUs, and IBM says it's a prime method for running common machine learning applications like Caffe, Torch, and Theano. While the Xeon Phi chip sits at the heart of Intel's machine learning hardware push, PowerAI uses IBM's Power8 processor with the Nvidia Tesla Pascal P100 GPU. Nvidia's Pascal line of GPUs is a next-generation machine learning server, due to a specifically tailored instruction set. But machine learning applications have to be written specifically to run that instruction set, and many cloud-based machine learning systems (like AWS) don't use Pascal hardware.