Category
NVIDIA, GPUs and AI chips: architecture, CUDA, the company and the industry from primary sources
Choosing AI infrastructure means reading a spec sheet correctly: sparse versus dense figures, compute versus memory bandwidth, theoretical peak versus measurement. This category uses NVIDIA whitepapers, SEC filings and peer-reviewed papers as primary sources to cover why GPUs are good at AI, how the company is governed and how the industry fits together. Values derived by calculation are labelled as theoretical, and vendor figures are kept apart from independent measurements.
9 articles in total
Foundational guide
Foundational guide (start here)
Why NVIDIA GPUs are so strong at AI — Tensor Cores, low precision, HBM and NVLink, read from formulas and published specs
Why NVIDIA GPUs are strong at AI, in four layers: Tensor Cores, low-precision formats (FP8/FP4), HBM and NVLink. Why the H100's 3,958 FP8 TFLOPS assumes sparsity, and why LLM decode is set by memory bandwidth, not compute — derived as theory from specs.
Related practical articles
- AIインフラ技術選定コスト最適化パフォーマンス
CPU vs GPU vs TPU: how they differ and how to choose — systolic arrays, memory bandwidth and cost per 1M tokens
What a TPU is, how it differs from CPUs and GPUs, why the TPU v1 "15-30x" figure is outdated, and a cost-per-token model that flips with batch size and SLOs.
42 min read - アーキテクチャ設計パフォーマンスインフラAI技術選定
How CPUs and GPUs really differ — pipelines, out-of-order execution, SIMT and the Roofline model, from formulas and measurements
CPU vs GPU from first principles: out-of-order execution versus SIMT and warps, with Roofline and PCIe math showing when moving work to a GPU pays off.
45 min read - AIパフォーマンスインフラアーキテクチャ設計技術選定
How CUDA works: a GPU programming primer — kernels, PTX/SASS, the memory hierarchy and synchronization, with complete code
How CUDA works: host and device, the grid/block/thread/warp hierarchy, NVCC→PTX→SASS, and GPU memory — with complete kernels and T4 vs H100 occupancy math.
63 min read - AI技術選定
Jensen Huang's life and technology leadership — his early years, career and the CUDA decision, checked against his own words and the record
Jensen Huang's background, career at AMD and LSI Logic, the NV1 failure, the CUDA bet and his 60-person staff, checked against his own words and filings.
30 min read - AI
Jensen Huang's resignation as NVIDIA CEO — separating confirmed facts, reporting and his own words, and what it says about founder dependence and governance
Jensen Huang resigned as NVIDIA's CEO on November 11, 2026, announced on the 12th. NVIDIA gave no reason; CFO Colette Kress is interim CEO. A timeline that separates SEC filings, reporting and his own words, plus the governance questions: a first-ever board chair and succession.
17 min read - AIインフラ技術選定コスト最適化アーキテクチャ設計
NVIDIA's Moat Is More Than a CUDA Monopoly — Breaking Its Advantage into Layers, and Reading Competitors and Future Scenarios from Primary Sources
NVIDIA's advantage layer by layer — CUDA, libraries, NVLink, supply — checked against ROCm, TPU and Trainium docs and SEC filings, with bull/base/bear scenarios.
36 min read - AIインフラ技術選定アーキテクチャ設計
A History of NVIDIA — From the 1993 Founding Through GeForce and CUDA to an AI Infrastructure Company, Read From Primary Sources
NVIDIA's history from primary sources, from NV1, GeForce and CUDA to Vera Rubin, and how Data Center grew from 4.8% to 89.7% of revenue in its 10-K filings.
34 min read - AIセキュリティインフラ技術選定
The US–China AI Chip War and Export Controls — Reading BIS's EAR, ECCN 3A090 and TPP from the Rule Text and Company Filings (as of November 17, 2026)
US export controls on AI chips to China, read from the rule text and NVIDIA's SEC filings: ECCN 3A090, TPP, the 2022–2026 rules, H20, H200 and Japan's controls.
37 min read