How the GPU Is Revolutionizing Machine LearningDecember 10, 2015
Machine learning is one of the most important computing developments of our time.
Advanced machine learning techniques are powering an explosion in artificial intelligence. They’re enabling new waves of smart applications and services.
Real-time speech translation. Autonomous robots. Detection of human emotions through facial analysis. These are just a glimpse of what’s possible.
But it takes a massive amount of computing performance to train the sophisticated deep neural networks that power these new applications. That’s a huge challenge. Training can take days to weeks on even the fastest supercomputers.
So it’s not surprising that virtually every leading machine learning researcher and developer is turning to our Tesla Accelerated Computing Platform and our Deep Learning software developers kit — a suite of tools and libraries for accelerating deep neural networks.
With GPU acceleration, neural net training is 10-20 times faster than with CPUs. That means training is reduced from days or even weeks to just hours. This lets researchers and data scientists build larger, more sophisticated neural nets, which leads to incredibly intelligent next-gen applications.
Leading Organizations Embrace GPUs for Machine Learning
Facebook is the latest organization to announce plans to use GPUs. Earlier today they unveiled “Big Sur” — their next-generation, purpose-built system for training deep neural networks.
Big Sur uses our new Tesla M40 GPU accelerators, which we designed to train deep neural networks in enterprise data centers. With GPUs, Big Sur is 2X faster than existing systems. It will allow Facebook to train twice as many neural networks. That boosts both application intelligence and accuracy.
And Facebook’s partners plan to open source the Big Sur specifications to the community via the Open Compute Project. This will let others harness GPUs for all types of machine learning work.
Other organizations are also using GPUs to power machine learning.
Last month, IBM announced that its Watson cognitive computing platform has added support for NVIDIA Tesla K80 GPU accelerators.
Added to Watson’s POWER architecture, GPUs accelerate its retrieve and rank API capabilities by 1.7X and deliver 10X greater processing power. This enhances Watson’s natural language processing capabilities and other key applications.
Advances in Machine Learning Powered by NVIDIA Deep Learning SDK
Key to the rapid adoption of GPUs for machine learning is NVIDIA’s Deep Learning SDK. It a suite of powerful tools and libraries that give data scientists and researchers the building blocks for training and deploying deep neural nets.
It includes DIGITS, NVIDIA’s Deep Learning GPU Training System. This lets data scientists and researchers quickly design the best deep neural network based on their data using real-time network behavior visualization. All without requiring them to write any code.
The SDK also includes cuDNN, the NVIDIA CUDA Deep Neural Network. Its optimized routines allow developers to focus on designing and training neural network models rather than low-level performance tuning.
It includes other libraries and tools as well — cuBLAS, cuSPARSE, NCCL and the CUDA toolkit — all optimized for machine learning workloads.
A Foundation for Deep Learning Frameworks
NVIDIA GPUs and the Deep Learning SDK are driving advances in machine learning. The number of organizations developing and using GPU-accelerated deep learning frameworks to train deep neural networks is growing.
These include Microsoft’s CNTK framework and Google’s TensorFlow. Their developers have recently made both available as open source solutions. They join other open frameworks — Caffe, Theano and Torch — that are widely used to design and train deep neural nets.
The AI race is on. And it’s powered by machine learning. As the go-to engine of machine learning, GPUs will help drive innovation in virtually all areas of industry and research.