Technology and Innovation Community

 View Only

  • 1.  NVIDIA $NVDA Keynote at COMPUTEX 2025

    Posted 31-05-2025 10:10

    Last week Jensen Huang, the Founder and CEO of NVIDIA $NVDA, delivered his annual keynote speech @ COMPUTEX 2025 in Taipei.

    Link to the event for those interested below:

    video

    #ai #innovationcommunity



    ------------------------------
    Todor Kostov
    Director
    ------------------------------


  • 2.  RE: NVIDIA $NVDA Keynote at COMPUTEX 2025

    Posted 02-06-2025 07:46

    Thanks Todor

    A detailed summary for those in a rush:

    1. AI Factories & Accelerated Computing Vision

    • Huang described AI factories as the new production facilities of the modern age, likening them to how past revolutions had textile mills or power plants.

    • These AI factories are data centers purpose-built to process, train, and infer AI models, transforming raw data into intelligence.

    • NVIDIA's full-stack offering includes:

      • GPU infrastructure (Blackwell and later Rubin architectures),

      • CUDA software platform, with massive ecosystem support,

      • Networking (InfiniBand and Spectrum-X), and

      • Inference orchestration platforms (like NIM).

    • The goal is to make AI development and deployment accessible, efficient, and scalable, enabling organizations of all sizes to run proprietary or open-source models.

    2. Rubin Platform

    • Unveiled as the successor to the Blackwell GPU platform, Rubin is expected to ship in 2026.

    • Key components:

      • Rubin GPU: Advanced GPU with support for HBM4 memory.

      • Vera CPU: Companion processor to the Rubin GPU for CPU-GPU co-optimization.

    • Highlights include better memory bandwidth, compute density, and integration of AI-specific workloads.

    • Rubin was presented as part of a predictable two-year cadence: Hopper → Blackwell → Rubin, ensuring developers and cloud providers can plan capacity expansions in sync with NVIDIA's roadmap.

    3. Networking Innovations

    • To address data movement bottlenecks in large-scale AI workloads, NVIDIA introduced enhancements in:

      • InfiniBand: Upgraded to support faster node-to-node connections with lower latency for AI workloads.

      • Spectrum-X: A new Ethernet-based networking platform optimized specifically for AI clouds, where traditional Ethernet often falls short.

    • Spectrum-X includes:

      • Spectrum-4 Ethernet switch

      • BlueField-3 DPUs (data processing units)

    • These innovations make AI cloud infrastructure more scalable, power-efficient, and reliable, especially for hyperscalers and large enterprises.

    4. AI Software Ecosystem

    • Huang emphasized that software is the "soul" of NVIDIA's platform, with a key focus on:

      • CUDA and cuDNN for foundational acceleration.

      • NIM (NVIDIA Inference Microservices): These containerized AI workloads simplify deploying powerful LLMs and vision models via APIs.

        • Example: Running LLaMA, Mistral, or OpenAI models directly through an enterprise-grade stack.

      • CUDA-Quantum: A hybrid programming model enabling developers to run classical and quantum workflows side by side.

    • The goal is to create a one-click deployment path for AI workloads, similar to how Kubernetes revolutionized container orchestration.

    5. Partnerships & Ecosystem

    • Major announcements included collaborations with:

      • Taiwanese OEMs (ASUS, Wistron, Pegatron, Foxconn, Quanta) to build new AI supercomputers and personal AI devices.

      • Global cloud providers (Google, Microsoft, AWS) for hosting NVIDIA AI stacks on their infrastructures.

      • Startups and research labs for genomics, robotics, and automotive.

    • Huang recognized Taiwan as the "engine room of the AI industrial revolution", critical in the design, manufacture, and scale-up of AI systems.

    6. AI Across Industries

    • Huang showcased real-world examples of NVIDIA's impact:

      • Biology: Using NVIDIA Clara and BioNeMo to simulate proteins and drug candidates.

      • Omniverse: Digital twins for warehouses, smart cities, and industrial robotics. One example showed a robot being trained in simulation and deployed in real time.

      • Healthcare: AI-assisted diagnostics and patient data processing for hospitals.

      • Automotive: NVIDIA DRIVE enabling software-defined vehicles with OTA updates and AV logic simulation in Omniverse.

    7. RTX AI PCs and Local AI

    • NVIDIA introduced a new category of RTX AI PCs, bringing generative AI to the personal computer:

      • Equipped with GeForce RTX GPUs and TensorRT software.

      • Local models like ChatRTX let users run custom LLMs (e.g., Mixtral or Gemma) using personal documents for contextual queries.

      • Tools for developers, gamers, and creators:

        • Image generation, real-time translation, content editing.

        • Integrations into productivity tools like OBS and generative fill in creative apps.

    • These PCs serve as entry-level AI factories, enabling low-latency AI inference without cloud dependence.

    8. Omniverse & Robotics

    • The Omniverse platform is evolving into a hub for simulating physical environments with real-world physics, AI agents, and robotics applications.

    • Huang demoed:

      • A digital twin of a warehouse where multiple robots trained via reinforcement learning perform pick-and-place tasks.

      • Use of generative AI to populate 3D scenes from simple prompts.

      • Interoperability with CAD systems, simulation engines, and sensor models.

    9. Closing Vision

    • Jensen concluded by reinforcing NVIDIA's role in shaping a world where every company becomes an intelligence producer.

    • He declared that NVIDIA is building the "engine room of the new industrial age"-AI factories that enable humanity to automate knowledge creation.

    • The future will be one of programmable industries, where software defines the logic, and AI powers the execution.



    ------------------------------
    Carlos Salas
    Portfolio Manager & Freelance Investment Research Consultant
    ------------------------------

    Original Message


  • 3.  RE: NVIDIA $NVDA Keynote at COMPUTEX 2025

    Posted 02-06-2025 08:54

    Great. Thanks Carlos.



    ------------------------------
    Todor Kostov
    Director
    ------------------------------

    Original Message


Related Content