Nvidia Goes Full-Stack, Touts Artificial Intelligence and Cloud

Nvidia is mapping out a customer service future populated with real-time avatars who use natural-language AI with real-world customers. The company, which has seemingly transformed from graphics powerhouse to AI authority (in just under 28 years since being founded by Jensen Huang, company CEO) used this week’s GTC conference to emphasize full-stack computing. The speed and flexibility of the company’s three GPU chips offer general purpose enterprise potential, thanks to Nvidia’s parallel-processing platform, CUDA. Huang backed this assertion with a slide indicating Nvidia has deployed more than 150 SDKs to industries generating $1 trillion.

In a nearly two-hour 2021 GTC keynote, Nvidia presented a vision of a world in which it facilitated virtual cars navigating virtual roads to design real-life autonomous vehicles, where test-runs with virtual robots resulted in improved physical robots and a world of additional improvements.

Nvidia showcased products including Metropolis, an edge-to-cloud platform for smart cities; Riva, an accelerated SDK for building AI speech applications; the open-source Merlin tools, that includes data sorting and warehousing and deployment of recommender systems; and to tie it all together, the Omniverse Avatar, a platform for generating AI avatars that “opens the door to the creation of AI assistants that are easily customizable for virtually any industry.”

An enterprise version of Omniverse starts at $9,000 per year. An enterprise version of Riva is expected to arrive in Q1 2022 (with a free version continuing to be offered to small firms and individual developers).

“Nvidia has created a ‘unified compute framework,’ which treats AI models as microservices that can be run together or in a distributed, hybrid architecture,” VentureBeat writes. “Nvidia’s concept for interoperable ‘metaverse’ virtual worlds” is built on the open-source Universal Scene Description (USD) specification created by Pixar to allow data from different source programs to be interoperable.

“We think of USD as the HTML of 3D,” Omniverse vice president Richard Kerris said in VentureBeat, explaining that, unlike HTML, USD is not governed the W3C (or a similar body) a consortium of companies is nonetheless working to advance it.

Key takeaways from Huang’s keynote:

  • Nvidia is offering a full-stack open platform at data center scale.
  • Accelerated computing, an Nvidia sweet spot, launched modern AI its resultant advancements.
  • Such advanced systems require three chips — the GPU, CPU and DPU (data processing unit) — which Nvidia has covered, “spanning from cloud to the edge.”
  • Huang introduced Nvidia Quantum-2, calling it “the most advanced networking platform ever built,” that with the Nvidia’s BlueField-3 DPU puts the company in the cloud-native supercomputing space.
  • Nvidia is building a simulation called Earth 2, our planet’s “digital twin,” to simulate and predict climate change.

Related:
Nvidia Makes Massive Language Model Available to Enterprises, VentureBeat, 11/9/21