NVIDIA GTC 2026: Everything Jensen Huang Just Announced (Full Breakdown) скачать в хорошем качестве

NVIDIA GTC 2026: Everything Jensen Huang Just Announced (Full Breakdown)

Jensen Huang just delivered the biggest GTC keynote in history — featuring actual keynote footage with Jensen speaking. $1 trillion in purchase orders, a record-breaking GPU, a $20 billion acquisition reveal, Disney robots, and AI data centers in space. Here's every major announcement broken down. 📑 CHAPTERS 0:00 $1 Trillion — The Biggest GTC Ever 0:47 Vera Rubin GPU — 336 Billion Transistors 1:58 NVL72 Rack — 3.6 Exaflops 3:14 Groq 3 LPU — The $20B Bombshell 4:42 Self-Driving — The ChatGPT Moment 5:45 Disney Olaf & 110 Robots 7:08 DLSS 5 — The Controversy 8:34 AI Data Centers in Orbit 9:26 DGX Station — 20 PFLOPS on Your Desk 10:22 Agentic AI & Open Models 11:30 Cloud Deals & Feynman 2028 12:52 The Numbers That Matter 🔥 KEY ANNOUNCEMENTS: VERA RUBIN GPU • 336 billion transistors (1.6x Blackwell) • 50 PFLOPS FP4 inference per chip • 288 GB HBM4 with 22 TB/s bandwidth • Dual-die, TSMC 3nm process NVL72 RACK • 72 Rubin GPUs + 36 Vera CPUs • 3.6 exaflops of FP4 compute • 1.3 million components per rack • Ships H2 2026 GROQ 3 LPU ($20B ACQUISITION) • 500 MB SRAM, 150 TB/s bandwidth • 35x more inference throughput per megawatt • 256 LPUs per rack, 40 PB/s bandwidth • Jensen called it "My Mellanox moment" SELF-DRIVING • "ChatGPT moment of self-driving" — Jensen • BYD, Hyundai, Nissan, Geely for Level 4 • Uber robotaxis in 28 cities by 2028 • Alpamayo AV model open-sourced on GitHub DISNEY OLAF ROBOT • Fully autonomous, no tethers/remote • Trained via deep reinforcement learning • Debuts Disneyland Paris March 29 DLSS 5 (CONTROVERSIAL) • Full neural rendering in real time • "GPT moment for graphics" — Jensen • Critics: "Snapchat filter for games" • Fall 2026 with Resident Evil, Hogwarts Legacy, Starfield SPACE COMPUTING • Vera Rubin Space-1: 25x H100 • Running LLMs directly in orbit • Partners: Axiom Space, Planet Labs, Starcloud DGX STATION • 20 PFLOPS desktop supercomputer • 748 GB coherent memory • Runs trillion-parameter models locally • Air-gapped option, six figures AGENTIC AI • NemoClaw: open-source agent platform • Nemotron Coalition with Mistral, Perplexity, Cursor • 6 model families across language, vision, robotics, driving, biology, weather FEYNMAN 2028 • 14x Blackwell performance • TSMC 1.6nm with silicon photonics • Scales to NVL1152 CLOUD PARTNERSHIPS • AWS: 1M+ NVIDIA GPUs • Azure: First to power NVL72 • Google Cloud: 76% cost reduction #NVIDIA #GTC2026 #JensenHuang #VeraRubin #Groq3 #DLSS5 #AI #GPU #SelfDriving #Robotics #Feynman #AINews #TechNews #ClaudeCode #AI #ChatGPT #ClaudeCode