Today on the podcast, we speak with Ian Buck and Kari Briski of NVIDIA about new updates and achievements in deep learning. Ian begins by telling hosts Jon and Mark about his first project at NVIDIA, CUDA, and how it has helped expand and pave the way for future projects in super computing, AI, and gaming. CUDA is used extensively in computer vision, speech and audio applications, and machine comprehension, Kari elaborates.

NVIDIA recently announced their new Tensor Cores, which maximize their GPUs and make it easier for users to achieve peak performance. Working with the Tensor Cores, TensorFlow AMP is an acceleration into the TensorFlow Framework. It automatically makes the right choices for neural networks and maximizes performance, while still maintaining accuracy, with only a two line change in Tensor Flow script.

Just last year, NVIDIA announced their T4 GPU with Google Cloud Platform. This product is designed for inferences, the other side of AI. Because AI is becoming so advanced, complicated, and fast, the GPUs on the inference side have to be able to handle the workload and produce inferences just as quickly. T4 and Google Cloud accomplish this together. Along with T4, NVIDIA has introduced TensorRT, a software framework for AI inference that’s integrated into TensorFlow.



Ian Buck

Ian Buck is general manager and vice president of Accelerated Computing at NVIDIA. He is responsible for the company’s worldwide datacenter business, including server GPUs and the enabling NVIDIA computing software for AI and HPC used by millions of developers, researchers and scientists. Buck joined NVIDIA in 2004 after completing his PhD in computer science from Stanford University, where he was development lead for Brook, the forerunner to generalized computing on GPUs. He is also the creator of CUDA, which has become the world’s leading platform for accelerated parallel computing. Buck has testified before the U.S. Congress on artificial intelligence and has advised the White House on the topic. Buck also received a BSE degree in computer science from Princeton University.

Kari Briski

Kari Briski is a Senior Director of Accelerated Computing Software Product Management at NVIDIA. Her talents and interests include Deep Learning, Accelerated Computing, Design Thinking, and supporting women in technology. Kari is also a huge Steelers fan.

Cool things of the week


Kubernetes 1.14: Production-level support for Windows Nodes, Kubectl Updates, Persistent Local Volumes GA blog
Stadia blog
How Google Cloud helped Multiplay power a record-breaking Apex Legends launch blog
Massive Entertainment hosts Tom Clancy’s The Division 2 on Google Cloud Platform blog


Interview


NVIDIA site
NVIDIA Catalog site
CUDA site
Tensor Cores site
TensorFlow sote
Automatic Mixed Precision for Deep Learning site
Automatic Mixed Precision for NVIDIA Tensor Core Architecture in TensorFlow blog
TensorFlow 2.0 on NVIDIA GPU video
NVIDIA Volta site
NVIDIA T4 site
WaveNet blog
BERT blog
Compute Engine site
T4 on GCP site
Webinar On Demand: Accelerate Your AI Models with Automatic Mixed-Precision Training in PyTorch site
PyTorch site
NVIDIA TensorRT site
TensorRT 5.1 site
Kubernetes site
Rapids site
NVIDIA GTC site
Deep Learning Institute site
KubeFlow Pipeline Docs site
KubeFlow Pipelines on GitHub site
NVIDIA RTX site


Question of the week

Where can we learn more about Stadia?


general info
developer access


Where can you find us next?

Mark will be at Cloud NEXT, ECGC, and IO.

Jon may be going to Unite Shanghai and will definitely be at Cloud NEXT, ECGC, and IO.

NVIDIA will be at Cloud NEXT and KubeCon, as well as International Conference on Machine Learning, The International Conference on Learning Representations, and CVPR