Mark Mirchandani and Jon Foust are together again this week, speaking with NVIDIA VP of Applied Deep Learning Research Bryan Catanzaro. Bryan and his team focus on using deep learning to enhance NVIDIA’s offerings.

Since Bryan was last a guest on the show, NVIDIA has been doing some amazing things. We talk about the A100 Tensor Core GPU and the massive effort it took to create, the new RTX graphics cards great for gaming, and the differences between them. Bryan explains how the new A100 chips compare to the previous versions, saying the new chips are larger, but with almost three times the power, making them ideal for things like precise calculations. And, as Bryan says, with better computation and more insight, we can make discoveries that benefit humanity. While the new RTX graphic cards are cheaper than previous versions, they are faster and more powerful, making gaming and video streaming much more enjoyable. Background noises and objects can even be removed with the help of deep learning.

Jon and Bryan talk about The Black Box at NVIDIA and what demos Jon hopes to see on his next visit. With this as the catalyst, Bryan talks more about how the NVIDIA architecture and the deep learning they employ have created efficient 8k graphics rendering for truly powerful gaming experiences. Outside of gaming, DLSS could have farther reaching benefits as the model learns new purposes, and Bryan talks us through some fun examples.

With the acquisitions of Mellanox and ARM, Bryan explains that NVIDIA has been able to streamline networking and really take advantage of powerful performance at all stages. The future of AI and HPC is about the data center, Bryan explains, and NVIDIA is hoping to push the boundaries on latency reduction and more.

Bryan Catanzaro

Bryan Catanzaro is VP of Applied Deep Learning Research at NVIDIA, where he leads a team finding new ways to use deep learning for graphics, speech, audio, and system design. His research led to the creation of the CUDNN library.

Cool things of the week


Introducing interactive code samples in Google Cloud documentation blog
GCP Podcast Episode 228: Fastly with Tyler McMullen podcast
Fastly Offers First Partner Edge Cloud-Based Content Delivery Solution on Google Cloud Marketplace press release and marketplace


Interview


NVIDIA site
NVIDIA A100 site
NVIDIA RTX 30 Series site
Mellanox site
ARM site
Cuda site
GCP Podcast Episode 119: NVIDIA and Deep Learning Research with Bryan Catanzaro podcast
GCP Podcast Episode 168: NVIDIA T4 with Ian Buck and Kari Briski podcast
Digital GTC site


Tip of the week

Zack Akil is here this week with a tip about ML on the edge with Teachable Machine and AutoML.

What’s something cool you’re working on?

Jon has been working on Open Match which went 1.0 recently! He’s been working on samples for matchfunctions, sending requests from game clients, and putting all that content out for the world to play with!