Although artificial intelligence in the client is all the rage, NVIDIA is focusing its latest edition of the AI ​​​​Decoded series on hybrid AI, a paradigm that is not yet familiar to most, but in fact aims to be by far the best solution for the many cases and needs in which we resort to artificial intelligence.

To understand exactly what hybrid AI consists of, we need to start from an important foundation, and that is the we are talking about using generative models to their full extentwhich, as you already know, is very broad and which at the same time aims to achieve in a more massive way solutions that aim to bring computing tasks to the local environment, which we have been witnessing for several months and which, with complete security, will intensify in the short and medium term .

So today the two most present paradigms are the following:

• AI in the cloud: all computing tasks are performed in the cloud using high-performance servers specialized for this type of tasks. It provides infinitely more computing capacity than the local system can offer.
• AI on campus– All computational tasks are performed in the system itself. It provides more privacy and for light tasks can significantly improve latency compared to a cloud service.

At this point, you’ve probably already guessed what hybrid AI consists of.

When we talk about hybrid AI, we mean an approach that proposes a combination of both paradigms, whereby we use both specialized hardware that we can rely on locally, and the processing capacity and scalability that AI platforms in the cloud offer us. And as I mentioned earlier, there are advantages and disadvantages to both, so we are faced with a model that suggests using the former and ignoring the latter.

NVIDIA can (and actually already does) focus on hybrid AI, given that it has been a major supplier of specialized AI components for years, whether for the domestic market with its RTX graphics adapters, or for professionals with specific solutions for workstations and similar environments. And this role will only increase in the future, as the company has been working on the IA PC Premium concept for some time, which is soon to be enhanced with its own SoC.

Now it may seem that we are talking about the future, but in reality Currently, we can already find the development of NVIDIA, which is based on hybrid artificial intelligenceand also for very diverse fields. Here are some of them:

• Generative AI from iStock: A collaboration between NVIDIA and the iStock platform, this solution allows us to conceive and prototype artwork in the cloud, then edit and finish it locally using tools accelerated by RTX hardware.
• NVIDIA ACE: This technology, which we told you about before, allows the interaction in games with NPCs (Non-Playable Characters) to be contextualized and therefore more natural and believable. It uses a hybrid AI model, combining local processing for minimal latency with cloud processing for more complex tasks.
• NVIDIA AI Workbench: Although the use of the optimized LLM is increasingly feasible locally, building, testing, and fitting the model is an extremely computationally intensive process, making it in many cases unavailable locally. This NVIDIA workspace offers the ability to move projects between on-premise and cloud based on performance and resource needs at any given moment.

Today, generative artificial intelligence is a transformative agent of the first orderbut limiting its use only at scale (on-premises or in the cloud) represents an important limitation that, as we have seen, will completely disappear when adopting a hybrid AI approach, a solution that will no doubt expand in the near future.