Zoom from a data center into an Nvidia Blackwell GPU die, down to a silicon atom: single-HTML 3D motion graphic by Opus 5.5 (X video, Korean)

Summary

This video showcases a 3D procedural motion graphic generated in a single HTML file by Anthropic's Claude Opus 5.5 (prompted at reasoning level xhigh). Shared by Korean creator 루샨 lucian (@lucian__03), the interactive WebGL/Three.js application smoothly zooms over 10 orders of magnitude from an entire AI data center room into an Nvidia Blackwell-era GPU, descending layer by layer down to individual silicon atoms in a transistor channel before zooming back out.

---

What is shown

• [00:00 - 00:02] Macro scale (Data center to server rack): An overhead perspective of a server hall with yellow overhead cable trays and rack rows (FOV ~8.97 m to 1.11 m), sliding out a conceptual generic server tray.
• [00:02 - 00:05] Board & Package level: Close-up of a dual-fan graphics card enclosure (FOV ~47.2 cm) opening up to expose the heat-spreader, GPU package die, GDDR7 memory modules, and substrate (FOV ~10.6 cm).
• [00:06 - 00:10] Silicon Die & Logical Blocks: The macroscopic floorplan of the die showing labeled L2 Cache (128 MB), GPCs, and individual Streaming Multiprocessors (SMs) containing 128 CUDA cores, 4 Tensor cores, and 1 RT core (FOV ~2.18 cm down to 1.43 mm).
• [00:11 - 00:16] Sub-block & Register File: Zooming into a single processing block showing warp schedulers, 64 KB register files, wordline decoders, and sense amplifiers (FOV ~304 µm down to 25.5 µm).
• [00:17 - 00:20] SRAM Memory Cell (6T): Close 3D view of a static RAM cell consisting of six transistors (cross-coupled inverters, pull-up, pull-down, access transistors) with illuminated yellow pulse lines demonstrating bitline/wordline readout (FOV ~551 nm).
• [00:21 - 00:25] Inside a CUDA core & Electron Microscope View: Functional layout of an FMA/ALU pipeline (INT32 unit, multiplier array, adder, alignment shifter) transitioning into a photorealistic simulated SEM view of polished transistor arrays (FOV ~53.8 µm down to 28.7 µm).
• [00:26 - 00:31] Transistor logic & FinFET architecture: Zooming into a full-adder logic tile (XOR cells), isolating a single FinFET gate wrapping around the silicon fin channel between source and drain (FOV ~1.55 µm down to 72.8 nm).
• [00:32 - 00:39] Atomic scale: The atomic lattice view highlighting the high-$k$ gate dielectric ($\text{HfO}_2$), $\text{TiN}$ gate metal, and individual silicon atoms with green glowing flowing conduction electrons spaced 0.235 nm apart at $1.96\text{ nm}$ field of view (labeled "6.3 Billion Times Closer").
• [00:40 - 00:48] Reverse zoom: A continuous camera fly-out retracing from the atomic lattice all the way back to the data center hall view (FOV 12.4 m).

---

Claims & numbers

• Zoom scale: Zooms from a 12.4-meter wide data center hall down to a 1.96-nanometer view of silicon atoms—an increase of approximately 6.3 billion times closer.
• Hardware specifications displayed in the graphic:
  • Full chip transistor budget: "90+ Billion Transistors".
  • L2 Cache: 128 MB between GPC banks.
  • Logical block layout: 188 SMs (based on RTX Pro 6000 spec table).
  • Per Streaming Multiprocessor: 128 CUDA cores, 4 Tensor cores, 1 RT core (total 24,576 CUDA cores on full chip).
  • Processing sub-block: 32 CUDA cores, 64 KB register file ($16,384 \times 32\text{-bit}$).
  • SRAM cell: 6 transistors (2 pull-up, 2 pull-down, 2 access).
  • Silicon atomic lattice spacing: $0.235\text{ nm}$ between adjacent bonded silicon atoms.

---

Notable quotes

• [00:00] "BLACKWELL: FROM THE DATA CENTER DOWN TO A SINGLE ATOM" (on-screen title).
• [00:21] "ONE CUDA CORE: 1 OF 128 PER SM · 24,576 ON THE FULL CHIP · THE VIEW IS ABOUT A HAIR'S WIDTH" (on-screen caption).
• [00:38] "6.3 BILLION TIMES CLOSER: THE VIEW IS NOW 1.96 NM WIDE — THE OPENING SHOT OF THE HALL WAS 12.4 M — THE ZOOM ENDS HERE" (on-screen caption).

---

Assessment

This is a screen recording of an interactive, code-rendered 3D visualization running locally from a browser (file:///Users/...), generated as a standalone WebGL script by Claude Opus 5.5. The visual transitions and scientific labeling are smoothly integrated and accurate to architectural schematics, serving as an educational motion graphic demonstration of LLM frontend and 3D rendering capabilities.

---

Lyrics & themes

• Audio/Narration: The video is purely visual (silent / ambient), with no vocal song lyrics or spoken voiceover.
• Themes: The visualization explores spatial scale, computer architecture hierarchy, and physical limits of silicon semiconductor engineering—linking warehouse-scale AI computing infrastructure to atomic quantum mechanics.

---

Lore & references

• Blackwell / RTX Pro 6000: Directly references Nvidia’s Blackwell GPU architecture, mapping out real component structures like GPCs, SMs, Tensor cores, and high-$k$ metal gates ($\text{HfO}_2$/$\text{TiN}$).
• "Claude Opus 5.5 (xhigh)": Highlighted in the top browser tab/header, demonstrating the model's high-effort reasoning capacity to generate complex, multi-thousand-line single-page procedural Three.js/WebGL applications without external assets.

---

Visual style & craft

• Rendering & Pipeline: Rendered entirely inside a standard web browser viewport using WebGL/Three.js primitives, shaders, dynamic LOD (Level of Detail), and a logarithmic camera zoom controller.
• Graphic Elements: Minimalist technical HUD styling with a real-time field-of-view (FOV) logarithmic scale bar on the right edge, dual English/Korean explanatory subtitles at the bottom, and bounding wireframe highlights pointing to active components.

Described by gemini-3.8-flash on 2026-10-05 from the video's audio and frames.