New AI Tools and Technologies Announced at GTC 2021 Keynote

Summary: NVIDIA’s GTC 2021 keynote unveiled a range of new AI tools and technologies designed to enhance AI development and deployment. Key announcements include NVIDIA Riva for conversational AI, NVIDIA Triton Inference Server for scalable AI, and NVIDIA DeepStream for video analytics. These tools aim to simplify AI development, improve performance, and enable real-time AI applications.

NVIDIA Unveils New AI Tools and Technologies at GTC 2021

NVIDIA’s GTC 2021 keynote was a showcase of the company’s latest advancements in AI technology. CEO Jensen Huang took the stage to introduce a range of new tools and technologies designed to empower developers and accelerate AI adoption.

NVIDIA Riva: Revolutionizing Conversational AI

NVIDIA Riva is a new software tool that enables developers to build engaging and accurate conversational AI applications. With Riva, enterprises can create AI agents that can understand and respond to customer queries in real-time. Key features include:

• Automatic Speech Recognition (ASR): Highly accurate speech recognition capabilities
• Real-Time Translation: Support for multiple languages and real-time translation
• Text-to-Speech: Expressive conversational AI agents

Riva is designed to help businesses improve customer engagement and reduce support costs.

NVIDIA Triton Inference Server: Scalable AI

NVIDIA Triton Inference Server is an open-source inference-serving software that enables fast and scalable AI deployment. Key features include:

• Model Analyzer: Determine optimal model execution parameters
• RAPIDS Forest Inference Library (FIL): Support for tree-based models
• Multi-GPU Multinode Distributed Inference: Support for giant Transformer-based language models
• Cloud Support: Available in all major public clouds

Triton Inference Server is designed to help developers deploy AI models quickly and efficiently.

NVIDIA DeepStream: Video Analytics

NVIDIA DeepStream is a powerful AI streaming analytics toolkit for building high-performance video analytics applications. Key features include:

• Low-Code Programming: Graphical programming with Graph Composer
• Action Recognition: Support for action recognition
• Automatic Speech Recognition (ASR): Support for ASR
• Audio-Video Synchronization: Support for audio-video synchronization
• Multiobject Tracker: Brand new multiobject tracker based on DeepSORT algorithm

DeepStream is designed to help developers build reliable AI applications faster.

NVIDIA AI Face Codec: Efficient Video Compression

NVIDIA AI Face Codec is a novel AI-based method for compressing videos and rendering human faces for video conferencing. Key features include:

• Up to 10x Reduction in Bandwidth: Compared to H.264
• Real-Time Rendering: Support for real-time rendering

AI Face Codec is designed to improve video conferencing efficiency and reduce bandwidth costs.

NVIDIA NeMo Framework: Large-Scale Language Models

NVIDIA NeMo Framework is a new capability for developing large-scale language models (LLM). Key features include:

• Automated Data Curation: Formatting, deduplication, and blending
• Advanced Parallelization Techniques: Pipeline, tensor, and data parallelism
• Scalable Training: Train models with trillions of parameters
• Customizable LLM: Train Megatron 530B for new domains and languages

NeMo Framework is designed to help enterprises train and scale language models efficiently.

Conclusion:

NVIDIA’s GTC 2021 keynote showcased a range of new AI tools and technologies designed to enhance AI development and deployment. From conversational AI to video analytics, these tools aim to simplify AI development, improve performance, and enable real-time AI applications. With these advancements, NVIDIA is poised to accelerate AI adoption across industries.

Table: Key Features of NVIDIA AI Tools