NVIDIA Corporation - Special Call - NasdaqGS:NVDA

NasdaqGS:NVDA

Paresh Kharya, Director of Product Management and Marketing at NVIDIA, provided a detailed overview of NVIDIA's recent product announcements and strategic directions. Here’s a summary of the key points he discussed:

Introduction

• NVIDIA recently held its GPU Technology Conference (GTC) digitally, which attracted 60,000 registered attendees, compared to the typical 10,000 in-person attendees.
• The keynote by Jensen Huang, NVIDIA's CEO, has been viewed 8 million times.

Trends Driving AI Adoption

• Exponentially Higher Compute Demand: AI models are becoming more complex, requiring significantly more compute power. For example, the NVIDIA Megatron-BERT model, which is for reading comprehension, requires 3,000 times more compute than the ResNet-50 model, which was the state-of-the-art when NVIDIA launched its Volta GPUs.
• Pervasive Use Cases: AI applications are now used in various domains, such as conversational AI, recommendation systems, and medical diagnostics. However, each query requires only a small amount of acceleration.

Fragmentation of Data Centers

• Current data centers are fragmented, with separate clusters for storage, CPU servers, AI training, inference, and general workloads.
• This fragmentation leads to inefficiencies due to unpredictable demand and underutilization.

Fungible Data Center Architecture

• To address the fragmentation issue, NVIDIA aims to create a "fungible data center" that can scale up for AI training and scale out for inference and other workloads, adapting dynamically to changing demands.
• This requires reimagining the GPU architecture, which led to the development of the NVIDIA Ampere GPU.

NVIDIA Ampere GPU (A100)

• Key Breakthroughs:
  • 20x higher performance compared to the previous Volta architecture.
  • Unified AI training and inference acceleration.
  • Massive scalability, allowing a single server to scale up as one giant GPU or scale out to 50 different accelerators.
• Technological Advancements:
  • 54 billion transistors, making it the world's largest 7-nanometer chip.
  • High-bandwidth memory (HBM2) with 1.5 terabytes per second of memory bandwidth.
  • Third-generation Tensor Cores with sparsity acceleration for higher performance.
  • Multi-Instance GPU (MIG) technology for elastic data centers.
  • Third-generation NVLink and NVSwitch for high-speed interconnectivity.

Performance Improvements

• A100 provides 6x higher performance for AI training and 7x higher performance for AI inference compared to Volta.
• BERT, an advanced AI model for natural language processing, showcases these performance improvements.

DGX A100 System

• DGX A100 is the world's first 5-petaflop server, featuring 8 A100 GPUs.
• It can be partitioned for multiple developers to work simultaneously.
• It is more compact and flexible than previous systems.

NVIDIA EGX Products

• For edge computing, NVIDIA introduced the next generation of EGX products, combining A100 GPU acceleration with Mellanox ConnectX-6 Dx for secure and efficient edge AI applications.
• Supports industries such as smart cities, healthcare, telecommunications, conversational AI, and robotics.

Conclusion

• Paresh highlighted the significant advancements in NVIDIA's technology and the potential impact on various industries, emphasizing the importance of accelerated computing and AI in driving innovation and efficiency.

Simona Jankowski, identified as the Former Vice President of Investor Relations, made the following statements during the NVIDIA special call:

1. Introduction and Welcome:

   • She thanked everyone for joining the call and welcomed attendees.
   • She mentioned that the call would be moderated by Evercore ISI semiconductor analyst C.J. Muse.

2. Webcast Details:

   • The call was being webcast live on NVIDIA's Investor Relations website.
   • The webcast would be available for replay until May 28, 2021.
   • She reminded participants that the content of the call is NVIDIA's property and cannot be reproduced or transcribed without prior written consent.

3. Forward-Looking Statements Disclaimer:

   • She noted that the call might contain forward-looking statements based on current expectations and that actual results may differ materially.

   • She advised listeners to refer to the disclosure in the previous earnings release, the most recent Forms 10-K and 10-Q, and any filings on Form 8-K with the SEC for factors that could affect future financial results and business.

   • She stated that all statements were made as of May 28, 2020, and based on information currently available to the company, and NVIDIA assumes no obligation to update any such statements except as required by law.