WWC24 - Ankit Patel - Unlocking the Future Breakthrough Application Performance and Capabilities with NVIDIA

A vision model sees a bus. An LLM interprets it. Another model speaks an alert. Learn how to compose specialized AI into powerful, real-time applications.

#1about 3 minutes

Understanding accelerated computing and GPU parallelism

Accelerated computing offloads parallelizable tasks from the CPU to specialized GPU cores, executing them simultaneously for a massive speedup.

#2about 2 minutes

Calculating the cost and power savings of GPUs

While a GPU-accelerated system costs more upfront, it can replace hundreds of CPU systems for parallel workloads, leading to significant cost and power savings.

#3about 4 minutes

Using NVIDIA libraries to easily accelerate applications

NVIDIA provides domain-specific libraries like cuDF that allow developers to accelerate their code, such as pandas dataframes, with minimal changes.

#4about 3 minutes

Shifting from traditional code to AI-powered logic

Modern AI development replaces complex, hard-coded logic with prompts to large language models, changing how developers implement functions like sentiment analysis.

#5about 3 minutes

Composing multiple AI models for complex tasks

Developers can now create sophisticated applications by chaining multiple AI models together, such as using a vision model's output to trigger an LLM that calls a tool.

#6about 2 minutes

Deploying enterprise AI applications with NVIDIA NIM

NVIDIA NIM provides enterprise-grade microservices for deploying AI models with features like runtime optimization, stable APIs, and Kubernetes integration.

#7about 4 minutes

Accessing NVIDIA's developer programs and training

NVIDIA offers a developer program with access to libraries, NIMs for local development, and free training courses through the Deep Learning Institute.