GPU-accelerated computing has fueled recent breakthroughs in artificial intelligence, scientific discovery and high performance computing. It now has a cloud computing pipeline to match.
We’ve worked with Amazon Web Services to create their newest and most powerful GPU-accelerated cloud offering: the AWS EC2 P2 instance. Researchers and data scientists can use the instance, which is powered by up to eight NVIDIA Tesla K80 data center GPUs, to accelerate a wide range of compute-intensive applications.
GPU-Powered Deep Learning Where Data Resides
GPUs are helping the world’s cloud customers tackle the explosion of data generated every day by transactional records, sensor logs, images, videos and more. GPU-powered deep learning makes it possible to process and generate insights from this data — and deliver intelligent, personalized experiences to customers.
For computers to understand speech with superhuman accuracy, interact in natural conversations andperform complex tasks safely and autonomously, they have to process immense amounts of data. Using AWS EC2 P2 instances, businesses and researchers can use the power of GPU-accelerated deep learning without the need to move data into and out of the cloud.
They can provide the most relevant experiences for their users, with the latest data. And they can use the latest deep learning models, which require exponentially higher compute power to process.
Fast Access to Powerful HPC Clusters
The AWS EC2 P2 instance also allows HPC customers looking to eliminate long job queues, meet their computing demands with a low upfront investment, or add short-term bursts of compute power to match peak workloads. For the most demanding HPC applications, up to 16 physical GPUs per instance and multiple instances within a placement group can provide the needed compute power.
To make it easy for people to get started with deep learning, NVIDIA has worked to ensure that every deep learning framework is accelerated on GPUs. In the same spirit, AWS is launching a new set of Amazon Machine Images with popular frameworks pre-installed, including MXNet, Caffe, Theano, TensorFlow and Torch.
To enable provisioning of GPU-accelerated HPC clusters in minutes rather than days or weeks, NVIDIA is providing AMIs in the AWS Marketplace preloaded with NVIDIA drivers, CUDA Toolkit and DIGITSdeep learning training software. Automation frameworks for deploying and maintaining HPC instances, like AWS’s CfnCluster framework, eliminate the need for applications to wait in job queues.