NVIDIA General Director Jensen Juang envisions a future where corporate software is revolutionized by the integration of chat bots into a unified system to handle tasks efficiently. Speaking at the GTC conference, Juang emphasized that the future of software development will not require writing code from scratch. Instead, teams of AI agents will be assembled, including models for task delegation, data analysis, and results presentation.
To enable the integration of various NVIDIA AI models, a new system called NVIDIA Inference Microservices (NIM) has been developed. NIM utilizes containers containing models and all necessary dependencies for deployment in any environment, including NVIDIA nodes.
Subscription to Nvidia AI Enterprise, priced at $4500 per year per GPU or $1 per hour per GPU in the cloud, is required for utilizing NIM. This innovation aims to simplify model packaging and optimization, ensuring high performance through regular updates.
In addition to optimization, NVIDIA is focused on streamlining communication between containers to facilitate seamless interaction among various models, despite differences in API implementation.
NVIDIA is not only striving to simplify AI usage for enterprises but also actively creating tools to customize models according to specific customer needs. Among these are nemo microservices for data preparation and NEMO RETRIEVER for integrating information from vector databases. While NIM and NEMO RETRIEVER models are already available, NEMO Microservices are currently in the early access stage.