I am excited to share with you that DeepLearning4J (DL4J), a Java-based open-source, distributed deep-learning toolkit for building, training and deploying Neural Networks is now available on Power. Following our foundational belief in collaborative innovation that inspires our work on the PowerAI ecosystem at large, the story of DeepLearning4J on Power is one of collaboration and hardware/software innovation, creating a system that combines best of breed hardware systems with innovative software technology.
Built on Java and distributed computing technologies, DL4J integrates naturally with Spark and supports training neural networks on a Spark cluster, in order to accelerate neural network training. Given the compute-intensive nature of Deep Learning workloads, could there be a better platform for DL4J than Power as the fastest Spark platform?
Our collaboration with Skymind to bring DL4J to Power began about a year ago when the PowerAI team first ported and optimized DL4J for Power. Since then, Skymind and the PowerAI team have worked together to streamline the build and installation process for DL4J and take advantage of the latest hardware and software enhancements for deep learning on Power, such as the vector-scalar floating point units of the POWER8 processor and the NVLink-attached GPUs introduced in POWER8+.
DeepLearning4J for Power is available directly from the DeepLearning4J repository and are installed with the Maven build tools. Read more about installation process. The nd4j-native-platform backend uses Power’s vector-scalar floating point units with the optimized OpenBLAS library for POWER8, and the nd4j-cuda-8.0-platform backend takes advantage of the most advanced GPU accelerators and high-performance NVLink interconnect to accelerate your Deep Learning workloads.
If you have questions about configuring and using DeepLearning4J to kickstart your Deep Learning deployment on Power, join the DeepLearning4J community on the Deeplearning4j Gitter channel!
Get started with Deep Learning to develop cognitive applications on Power! Share how you are unleashing the power of deep learning to transform the future of computing in the comments section.
Dr. Michael Gschwind is Chief Engineer for Machine Learning and Deep Learning for IBM Systems where he leads the development of hardware/software integrated products for cognitive computing. During his career, Dr. Gschwind has been a technical leader for IBM’s key transformational initiatives, leading the development of the OpenPOWER Hardware Architecture as well as the software interfaces of the OpenPOWER Software Ecosystem. In previous assignments, he was a chief architect for Blue Gene, POWER8, POWER7, and Cell BE. Dr. Gschwind is a Fellow of the IEEE, an IBM Master Inventor and a Member of the IBM Academy of Technology.