As Nvidia pushes to democratize AI, right here’s all the things it introduced at GTC 2023

At this yr’s GPU Expertise Convention (GTC), Nvidia continued its AI {hardware} push with a selected deal with making its know-how extra accessible to enterprises throughout industries and streamlining the event of generative AI purposes like ChatGPT.

The next is a each day recap of main bulletins that the Santa Clara, California-based firm made with hyperlinks to in-depth protection.

Hire AI supercomputing infrastructure with DGX Cloud

Whereas Nvidia has been constructing {hardware} for AI for fairly a while, the know-how has taken a while to see mass adoption — partly owing to excessive prices. Again in 2020, its DGX A100 server field was bought for $199,000. To alter this, the corporate at present introduced DGX Cloud, a service that may enable enterprises to entry its AI supercomputing infrastructure and software program by an online browser.

>>Comply with VentureBeat’s ongoing Nvidia GTC spring 2023 protection<<

DGX Cloud rents DGX Server packing containers, every with eight Nvidia H100 or A100 GPUs and 640GB of reminiscence, and prices $36,999 a month for a single node.

Leveraging the facility of DGX Cloud, the corporate additionally introduced the launch of AI Foundations to assist enterprises create and use customized generative AI fashions. The providing, Nvidia stated, gives three cloud providers: Nvidia NeMo for big language fashions (LLMs), Nvidia Picasso for picture, video and 3D purposes, and BioNeMO to generate scientific texts primarily based on organic information.

New {hardware} for AI inference and suggestions

Alongside DGX and AI Foundations, Nvidia additionally debuted 4 inference platforms designed to assist builders shortly construct specialised generative AI purposes. This consists of Nvidia L4 for producing AI video; Nvidia L40 for 2D/3D picture technology; Nvidia H100 NVL for deploying giant language fashions; and Nvidia Grace Hopper — which connects the Grace CPU and Hopper GPU over a high-speed 900GB/sec coherent chip-to-chip interface — for advice techniques constructed on big datasets. 

The corporate says L4 can ship 120x extra AI-powered video efficiency than CPUs, mixed with 99% higher vitality effectivity; whereas L40 serves because the engine of Omniverse, delivering 7x the inference efficiency for Secure Diffusion and 12x Omniverse efficiency over the earlier technology.

Chipmakers get cuLitho at Nvidia GTC

On the occasion, Nvidia CEO Jenson Huang took the stage to announce Nvidia cuLitho software program library for computational lithography. The providing, as Huang defined, will allow semiconductor enterprises to design and develop chips with ultrasmall transistors and wires whereas accelerating time to market and boosting the vitality effectivity of the large information facilities that run 24/7 to drive the semiconductor manufacturing course of.

“The chip business is the inspiration of almost each different business on this planet,” stated Huang. “With lithography on the limits of physics, NVIDIA’s introduction of cuLitho and collaboration with our companions TSMC, ASML and Synopsys permits fabs to extend throughput, cut back their carbon footprint and set the inspiration for 2nm and past.”

Lastly, the corporate additionally introduced partnerships with Medtronic and Microsoft. The previous, it stated, will result in the event of a standard AI platform for software-defined medical units able to enhancing affected person care. In the meantime, the latter will see Microsoft Azure host Nvidia Omniverse and Nvidia DGX Cloud.

The 2023 Nvidia GTC occasion runs by March 23.