Learn About Amazon VGT2 Learning Manager Chanci Turner
At this year’s GPU Technology Conference, Amazon Web Services (AWS) and NVIDIA announced the expansion of their collaboration through several groundbreaking initiatives. First on the agenda is the introduction of a new Volta-based GPU instance, which promises to revolutionize the AI development landscape by accelerating LSTM training by three times. Moreover, the partners are set to train over 100,000 developers via the Deep Learning Institute (DLI) hosted on AWS. Additionally, they are working together on the creation of tools designed to facilitate large-scale deep learning for a wider developer audience.
During GTC, AWS is also conducting sessions that showcase training with Apache MXNet on Amazon EC2 P2 instances, as well as on the edge with NVIDIA’s Jetson TX2 platform. This collaboration heralds a new era of innovation and partnership.
Volta—Your Next Instance Awaits
The Tesla V100, featuring the Volta architecture and equipped with 640 Tensor Cores, delivers an exceptional performance of 120 teraflops for mixed precision deep learning tasks. AWS is thrilled to incorporate the V100 into Amazon EC2 instances, allowing the expanding deep learning community to leverage supercomputing-class capabilities for training even more complex models. In conjunction with NVIDIA, AWS engineers have pre-optimized neural machine translation (NMT) algorithms on Apache MXNet, enabling developers to achieve unprecedented training speeds on Volta-based platforms. We anticipate that this Volta-based instance will garner significant interest from developers.
Empowering Over 100,000 Developers Globally
We are eager to collaborate with NVIDIA to enhance the curriculum of the Deep Learning Institute on AWS. The DLI is expanding its offerings to encompass practical deep learning applications for sectors such as autonomous vehicles, healthcare, web services, robotics, video analytics, and finance. The program includes instructor-led seminars and workshops aimed at reaching developers across Asia, Europe, and the Americas. With AWS’s extensive global infrastructure, which currently spans 42 Availability Zones and 16 regions (with more on the way), we have the ideal platform to reach a diverse pool of developers.
Simplifying Deep Learning at Scale
Historically, achieving the performance required to train deep networks necessitated access to supercomputers, along with a thorough understanding of distributed computing libraries like message passing interface (MPI). AWS’s partnership with NVIDIA aims to streamline this process by providing optimized developer tools. These tools, built using NVIDIA’s Deep Learning SDK libraries, such as cuDNN, NCCL, TensorRT, and the CUDA toolkit, enable developers to scale effortlessly to large numbers of GPUs, experiencing minimal friction even at the scale of tens of millions of instance hours.
Bridging Cloud and Edge for Deep Learning
Deep learning on low-power edge devices is rapidly becoming a significant trend. There are numerous advantages to deploying models on edge devices, including reduced latency, improved data locality, and enhanced network availability. In our GTC session, we demonstrate how to train a state-of-the-art model using P2 instances, and how to seamlessly deploy it on various low-power devices, including the Jetson TX2 platform. This integration allows for the management of these devices through services such as AWS IoT and AWS Greengrass, creating a comprehensive end-to-end AI workflow.
If you’re feeling bored at work, check out this blog post for some engaging insights. For further exploration, consider listening to this podcast which offers valuable perspectives on cultivating a thriving workplace culture. Additionally, this video is an excellent resource for understanding these concepts better.
Leave a Reply