A translation widget is provided for your convenience to facilitate translation of the English language version of this blog into several languages. If you choose to utilize this automated translation facility, please understand there may be deviations between the automated translation and the original English version. IBM is not responsible for any such automated translation deviations and offers the translated version "AS IS" without warranties of any kind.
IBM Power Systems: The leading platform for deep learning–based AI for the enterprise
Picture this: A few years ago, the CEOs of every major enterprise in the world leaned over to their CIOs and CTOs and said, “We have to figure out big data.” Recently, they asked their teams to embrace the cloud. The strategic imperative this year is artificial intelligence (AI).
Enterprises are looking to new AI methods to garner insights from data, to offer customers a better experience and to maximize revenue. From better fraud detection to chatbots to recommendation engines, enterprises are fast adopting a new data analytics method called deep learning. Deep learning is used in many everyday mobile services we use, such as Google Voice and Apple Siri, to enhance our interactions with our smartphones.
These deep learning–based AI methods work by taking in lots of data and training a computer how to extract insight from it. This requires significant computing power, and GPU-accelerated servers have become the norm for running AI workloads.
Lessons from the High-Performance Computing community
The world of high-performance computing has long recognized the importance of scalable, differentiated hardware that pushes the boundaries of computing. That’s why so many supercomputing clusters around the world run on IBM Power Systems. It’s also why Power Systems received two of the US Department of Energy’s CORAL grants. When Oak Ridge and Lawrence Livermore National Laboratories selected IBM to build their new supercomputers, Summit and Sierra, they did so for the Power architecture’s data-centric computing and powerful GPU and FPGA accelerator technology.
As GPU-driven cognitive computing continues to grow, so does the advantage of IBM Power Systems. With the first-ever NVIDIA NVLink GPU-to-CPU interface, the IBM S822LC for HPC with four NVIDIA P100 Tesla GPUs outperforms a competitor’s x86 system by 2.5 times! This enables researchers to tackle a wide variety of problems from climate change to aeronautical engineering in new ways. You can see more about how Oak Ridge is using IBM Power Systems to solve these challenges.
But different problems require different tools, and that’s why IBM Power Systems also feature the Coherant Accelerator Processor Interface, or CAPI. CAPI allows researchers to program FPGA cards that are deeply integrated with the Power CPU, offering faster, simpler FPGA analysis. With this technology, the University of Auburn is using CAPI-attached FPGAs and POWER8 to combat the growing problem of malware on Android devices, again highlighting the versatility of the Power architecture for research applications.
The HPC community was ahead of the game when it came to cognitive, already exploring deep learning and machine learning for cognitive solutions to the world’s toughest problems. However, as the promise of cognitive became more and more enticing, the business world also began taking notice of IBM Power Systems for the next frontier of computing.
Transforming businesses with high-performance data analytics
The best practices of HPC systems using accelerator technology have infiltrated the boardroom as IBM Power Systems helps business clients of all sizes achieve competitive differentiation and explore new cognitive technology. No longer doing just high-performance computing, businesses have adopted high-performance data analytics to find better AI-driven ways to help their customers.
One such business is Kinetica, a GPU-accelerated database provider. Powered by the IBM NVLink server, Kinetica helps major retailers deliver hyper-targeted, real-time marketing offers to their customers, bringing them the information they need to make a purchase, at the time they want to make the purchase. The Kinetica GPU database performs 2.5x better on POWER8 with NVLink systems, and is helping to usher in “real real-time” analytics for retailers and more.
Embracing the power of open for deep learning with PowerAI
As founding members of the OpenPOWER Foundation, IBM embraces collaborative innovation. Unlike some vendors, we realize that true deep-learning superiority can be achieved only by working with a community of experts, as evidenced by our integration of NVLink. That’s why we created PowerAI, the world’s leading open source, deep-learning toolkit for the enterprise. Unpacked through a simple binary, PowerAI brings optimized and integrated versions of the best open source, deep-learning frameworks, including the recently announced TensorFlow 1.0. Even better, with optimization and tuning for Power Systems, TensorFlow runs 30 percent faster on Power than on x86. Already we’re seeing users leveraging PowerAI in a variety of ways, such as improving worker safety with the use of drones, increasing accuracy in credit risk analysis and more.
With PowerAI, users are able to get off the ground and run much more simply, without a staff of PhDs to get their deep-learning network trained. Throw in the enterprise-class support from IBM and IBM Lab Services, and it’s easy to see why PowerAI is the preferred choice for deep learning.
Deep learning–based AI solutions provide a great opportunity for IBM Business Partners, and Power Systems is the leading platform for these solutions. You can let me know what you think by using the comments feature below.
VP, High Performance Computing & Data Analytics
Discover the new OpenPOWER LC servers: http://ibm.co/1RqkhYY
Help your customers get started with PowerAI: http://ibm.co/2ngVGkb
Tackle new problems with NVIDIA Tesla P100 on the only architecture with CPU:GPU NVLink: http://ibm.co/29AXauT