NVIDIA GTC 2024: Top 5 Trends

NVIDIA has travel a agelong mode from the days of specializing successful graphics cards for gaming — NVIDIA GPUs present supply a batch of the powerfulness down generative AI for enterprise. At NVIDIA GTC 2024, held March 18 – 21 successful San Jose, California, generative AI was everywhere, from chatbots to creation installations. Here are immoderate of the apical tech trends we saw portion astatine NVIDIA GTC this year, meaning the tech that came up implicit and implicit again successful presumption topics, the keynote and property Q&A with NVIDIA CEO Jensen Huang, and connected the amusement floor.

Retrieval-augmented generation

Billed arsenic a method for cutting down connected AI “hallucinations” oregon inaccuracies, retrieval-augmented procreation lets a generative AI exemplary cheque its enactment against outer resources specified arsenic probe papers oregon articles. RAG appeals to endeavor customers due to the fact that it increases the reliability of generated content.

SEE: NVIDIA CEO Jensen Huang revealed the upcoming Blackwell-architecture GPUs and much during the league keynote. (TechRepublic)

For example, Lenovo is an aboriginal adopter of NVIDIA’s recently announced NeMo model with RAG, which Lenovo is utilizing to physique retired its AI ecosystem for customers who enactment connected Lenovo devices.

“AI factories” for accrued retention and compute needs

Many organizations astatine NVIDIA GTC positioned themselves arsenic “AI factories,” which springiness enterprises entree to the retention and compute powerfulness they request to marque backstage AI.

NexGen Cloud, which calls its AI mill work “GPUaaS,” is among the companies that volition provide entree to NVIDIA’s 10-trillion parameter Blackwell GPU (Figure A) aboriginal this year.

Figure A: The Blackwell level and Blackwell-architecture GPUs were down galore of the innovations astatine NVIDIA GTC 2024. Image: NVIDIA

Ten trillion parameter jobs necessitate a batch of compute, and organizations are betting they tin marque a concern exemplary retired of providing conscionable the close magnitude of that computing powerfulness to customers.

“As those models get bigger and bigger, continuing to turn exponentially, the infrastructure that’s required to train, fine-tune and service oregon supply inference for those astatine standard besides needs to proceed to turn to lick that problem,” said Mark Lohmeyer, vice president and wide manager of compute and AI/ML infrastructure astatine Google Cloud, successful an interrogation with TechRepublic astatine NVIDIA GTC 2024.

Storage needs to enactment highly performant structured information arsenic good arsenic unstructured information specified arsenic documents, images and video, said Greg Findlen, elder vice president of merchandise absorption of information absorption astatine Dell, astatine a pre-briefing connected March 15. Customers besides privation to beryllium capable to negociate however their processes are utilizing the disposable hardware. “Nobody wants to person idle GPUs,” Findlen said.

The Dell AI Factory, developed with assistance from and enactment for NVIDIA products, is meant to constrictive down “vast possibilities” into “impactful usage cases,” said Varun Chhabra, Dell elder vice president of infrastructure and telecom marketing, astatine the pre-briefing.

According to a Gartner study published successful March 2024, 83% of 459 exertion work providers polled from October to December 2023 had deployed oregon were piloting generative AI wrong their organizations.

Edge AI

Organizations focusing connected borderline AI took up a ample information of the amusement level astatine NVIDIA GTC 2024, with a wide assortment of usage cases: robotics, automotive, industrial, warehousing, healthcare, captious systems and retail.

Many of these borderline AI usage cases were powered by NVIDIA’s Jetson level for robotics. NVIDIA Metropolis microservices connected Jetson Orin lets developers usage API calls to acceptable up generative AI capabilities connected the edge, making robots much reactive and flexible to their environments.

For example, during the keynote, NVIDIA CEO Jensen Huang showed a objection of warehouse robots that automatically rerouted astir an obstacle (Figure B).

Figure B: Robots and AI agents are trained successful a simulated concern abstraction utilizing a operation of NVIDIA software: Omniverse, Metropolis, Isaac and cuOpt. Image: NVIDIA

“AI is not new, but the speech astir generative AI is reinvigorating this taxable for many,” said Chhabra successful an email to TechRepublic. “We’ve been doing AI inferencing astatine the borderline for years, and information scientists person been utilizing our endpoints similar Dell Precision workstations to bash AI modeling and impervious of concept.”

Private AI for enterprise

Organizations are moving connected spinning up backstage generative AI that tin entree proprietary information securely portion providing the flexibility of a nationalist AI similar ChatGPT.

A communal sanction connected the amusement level for backstage AI services was Mistral AI, which provides an unfastened root ample connection exemplary that customers tin big connected their ain servers.

Copilots

Copilots aren’t caller — chatbots similar ChatGPT acceptable disconnected the generative AI boom, aft all. Since then, “copilot” has go astir a generic word for a chatbot that tin reply questions astir data.

Copilots tin gully from company-owned data

NVIDIA GTC saw a wide scope of copilot AI that tin gully answers from specific, company-owned structured and unstructured data. For example, the SoftServe Gen AI Industrial Copilot reads from a robot arm’s attraction manual to make step-by-step instructions for making repairs and tin item the parts the technician needs to regenerate connected a 3D model.

Citation lets humans cheque AI answers

Another communal inclination successful endeavor copilots was citation. NexGen Cloud showed however its Hyperstack unreality level (developed by SoftServe and accelerated by NVIDIA GPUs) could tally a copilot that could reply questions based connected a video and constituent backmost to circumstantial moments successful the video transcript wherever the AI sourced its answers. The operation of proprietary, backstage information sources with copilot-style chatbot functionality continues to beryllium the driving inclination successful generative AI for enterprise.

Disclaimer: NVIDIA paid for my airfare, accommodations and immoderate meals for the NVIDIA GTC lawsuit held March 18 – 21 successful San Jose, California.