GTC China 2019 Jensen Huang Keynote

Category: China Keynote · Year: 2019 · ▶ Watch

Speakers: Jensen Huang - Founder and CEO / NVIDIA · Nathan Ratliff - Lead Architect / NVIDIA · Rev Lebaredian - VP, Simulation Technology / NVIDIA

Switch language → 中文

Segments (9)

  • 00:00 · Introduction and Accelerated Computing
    • Jensen Huang introduces the conference, highlights the growth of GTC, and discusses the full-stack approach of NVIDIA’s accelerated computing.
  • 12:51 · Computer Graphics and RTX
    • Showcasing the impact of RTX ray tracing in gaming, including Minecraft and Bright Memory, and introducing Max-Q laptops.
  • 21:13 · Omniverse AEC
    • Introduction of Omniverse for Architecture, Engineering, and Construction, enabling collaborative, real-time photorealistic design.
  • 23:20 · Cloud Gaming
    • Announcement of Tencent’s START cloud gaming service powered by NVIDIA.
  • 24:18 · High Performance Computing
    • Discussing HPC applications including a Mars lander simulation, genomics sequencing with Parabricks, and bringing CUDA to the Arm architecture.
  • 31:00 · Artificial Intelligence and Recommender Systems
    • Exploring the evolution of AI from image recognition to natural language processing, and the massive scale of deep recommender systems used by Baidu and Alibaba.
  • 41:23 · TensorRT 7
    • Unveiling TensorRT 7, which adds support for RNNs and Transformers, enabling real-time conversational AI.
  • 45:30 · Autonomous Vehicles and DRIVE AGX Orin
    • Detailing the software-defined AV platform, federated learning, and announcing the next-generation Orin processor.
  • 59:30 · Robotics and Isaac SDK
    • Showcasing the Isaac robotics platform, simulation capabilities, and a live demonstration of the Leonardo robot.

Product Announcements (7)

  • [21:13] NVIDIA Omniverse AEC
    • A collaborative platform for architecture, engineering, and construction.
    • specs: Real-time ray tracing, multi-tool collaboration via USD, physically accurate rendering.
    • availability: Early access available
  • [23:20] Tencent START Cloud Gaming
    • A cloud gaming service powered by NVIDIA GPUs.
    • specs: Brings PC gaming to underpowered devices.
    • availability: Beta trials starting in China
  • [25:40] NVIDIA Parabricks on NGC
    • A genomics analysis toolkit accelerated by GPUs.
    • specs: Accelerates GATK pipeline, processes 60 genomes/day on BGI supercomputer.
    • availability: Available now on NGC
  • [26:31] CUDA on Arm
    • NVIDIA’s accelerated computing stack brought to the Arm CPU architecture.
    • specs: Supports HPC and AI workloads on Arm-based servers.
    • availability: Available now
  • [41:23] TensorRT 7
    • The 7th generation of NVIDIA’s deep learning inference compiler.
    • specs: Support for CNNs, Transformers, RNNs; over 1000 optimized kernels; enables conversational AI in under 300ms.
    • availability: Available soon on NGC
  • [48:14] NVIDIA DRIVE AGX Orin
    • Next-generation software-defined AV and robotics processor.
    • specs: 17 billion transistors, 12 Arm Hercules CPU cores, 200 TOPS, 7x performance of Xavier.
    • availability: 2022 SOP (Start of Production)
  • [1:01:40] NVIDIA Isaac SDK Updates
    • Software development kit for robotics.
    • specs: Includes Isaac Sim, Isaac Gym, and pre-trained models for navigation and manipulation.
    • availability: Available now

Specific Numbers (7)

Timestamp Metric Value Context
03:43 Attendees 6,100 Registered attendees for GTC China, a 250% growth in 3 years.
14:06 Active Players 100 million Monthly active players for Minecraft globally (300 million registered in China).
21:22 Laptops Sold 5 million Gaming laptops sold in China in the last year.
25:55 Genomes Sequenced 60 per day Throughput of the BGI MGI-T7 Life Science Supercomputer using Parabricks.
39:23 Cost $40 Cost to render a specific sequence on RTX GPUs, compared to $310 on CPUs.
48:14 Transistors 17 billion Number of transistors in the new Orin SoC.
48:14 Performance 200 TOPS Trillion operations per second delivered by the Orin SoC.

Benchmark Claims (5)

  • [10:38] AI Training Performance: 4x faster
    • vs: Previous software versions on the same hardware
    • gain: Quadrupled performance in 2 years purely through software optimization.
  • [10:58] AI Inference Performance: 2x faster
    • vs: Previous software versions on the same hardware
    • gain: Doubled performance in 1 year purely through software optimization.
  • [39:23] Rayvision Cloud Rendering: 39 hours / $40
    • vs: 485 hours / $310 on CPU
    • gain: Massive reduction in both rendering time and cost.
  • [36:00] Baidu Recommender Training: 1/10th the cost
    • vs: CPU-based training
    • gain: 90% cost reduction for training complex recommendation models.
  • [39:00] Alibaba Recommender Throughput: 100x higher
    • vs: Previous CPU implementation
    • gain: Two orders of magnitude increase in recommendation throughput.

Customer Stories (7)

  • [15:20] Minecraft (Microsoft/Mojang)
    • Integrated RTX path tracing into Minecraft.
    • outcome: Transformed the visual quality of the game with real-time lighting, shadows, and reflections.
  • [23:20] Tencent
    • Launched START cloud gaming service powered by NVIDIA.
    • outcome: Brings high-end PC gaming experiences to underpowered devices across China.
  • [25:40] BGI
    • Used NVIDIA Parabricks on their MGI-T7 supercomputer.
    • outcome: Achieved the ability to sequence 60 whole genomes per day.
  • [38:28] Rayvision
    • Adopted NVIDIA RTX for their cloud rendering platform.
    • outcome: Reduced rendering time from 485 hours to 39 hours, and cost from $310 to $40 for a specific workload.
  • [36:00] Baidu
    • Moved their deep recommender system to NVIDIA GPUs.
    • outcome: Reduced training cost by 90% (1/10th the cost of CPUs).
  • [39:00] Alibaba
    • Used NVIDIA AI for their Singles Day recommendation engine.
    • outcome: Achieved 10% higher Click-Through Rate (CTR) and 100x higher throughput.
  • [54:30] Didi
    • Selected NVIDIA DRIVE for autonomous driving and cloud infrastructure.
    • outcome: Building their next-generation robotaxi fleet using NVIDIA technology.

Key Technologies (7)

  • CUDA: NVIDIA’s parallel computing platform and programming model.
  • RTX: Hardware-accelerated ray tracing technology for photorealistic graphics.
  • Omniverse: A collaborative platform for 3D design and simulation.
  • Parabricks: GPU-accelerated genomics analysis software.
  • TensorRT: Deep learning inference optimizer and runtime.
  • DRIVE: End-to-end platform for autonomous vehicle development.
  • Isaac: Software-defined robotics platform including simulation and SDKs.

Demos Shown (5)

  • [15:20] Minecraft with RTX on and off.
    • True
  • [18:35] Bright Memory game with RTX.
    • True
  • [22:38] Omniverse AEC demonstrating real-time collaborative architectural design (KPF building).
    • True
  • [33:06] Mars lander simulation using fluid dynamics and RTX rendering.
    • True
  • [1:02:06] Leonardo robot live on stage, recognizing and stacking colored blocks.
    • True

Predictions / Commitments (3)

  • [17:50, Current and ongoing] Ray tracing is the future. RTX has been a home run. We have redefined computer graphics.
  • [55:14, Long-term future] Everything that moves will be autonomous someday.
  • [55:35, Near to long-term future] We are going to see the smart everything revolution.

Companies Mentioned (8)

Microsoft · Tencent · Autodesk · Arm · Ericsson · Baidu · Alibaba · Didi

Notable Quotes (2)

The more you buy, the more you save. — Jensen Huang @ 39:45

Everything that moves will be autonomous someday. — Jensen Huang @ 55:14

Key Topics

Accelerated Computing · CUDA · RTX Ray Tracing · Omniverse · Cloud Gaming · Genomics · Arm Architecture · Deep Recommender Systems · Conversational AI · TensorRT · Autonomous Vehicles · Federated Learning · Robotics · Isaac SDK

Takeaways

  • NVIDIA’s full-stack approach to accelerated computing continues to yield massive performance gains purely through software optimization.
  • RTX ray tracing is becoming the standard for computer graphics, expanding beyond gaming into architecture and film.
  • NVIDIA is expanding its reach into new architectures by bringing CUDA to Arm.
  • Deep recommender systems are the ‘engine of the internet,’ and GPUs offer massive cost and throughput advantages over CPUs for these workloads.
  • TensorRT 7 enables real-time conversational AI by optimizing complex models like RNNs and Transformers.
  • The new DRIVE AGX Orin processor provides a 7x performance leap for autonomous vehicles, maintaining software compatibility.
  • NVIDIA is building an end-to-end ecosystem for robotics with the Isaac platform, emphasizing simulation and pre-trained models.