Archive - SemiAnalysis

Nvidia Blackwell Perf TCO Analysis - B100 vs B200 vs GB200NVL72

GPT-4 Profitability, Cost, Inference Simulator, Parallelism Explained, Performance TCO Modeling In Large & Small Model Inference and Training

Apr 10 •

and

Daniel Nishball

Is Intel Back? Foundry & Product Resurgence Measured

Foundry Cost Wall, Whale Customers, Datacenter Share, The Money Problem

Apr 2 •

,

,

Daniel Nishball

,

Jeremie Eliahou Ontiveros

, and

March 2024

Nvidia’s Optical Boogeyman – NVL72, Infiniband Scale Out, 800G & 1.6T Ramp

Transceiver to GPU Ratio, DSP Growth, Revealing The Real Boogeyman

Mar 25 •

and

Daniel Nishball

Nvidia B100, B200, GB200 - COGS, Pricing, Margins, Ramp - Oberon, Umbriel, Miranda

The B Stands For Jensen's Benevolence

Mar 18 •

,

, and

Astera Labs IPO - The Next Connectivity Superhero or Steamrolled By Competition?

Bottoms Up Model, Units, ASP, Revenue, By Hyperscaler Analysis, EPS & Cashflow, Competitive Analysis

Mar 17 •

and

Jeremie Eliahou Ontiveros

CXL Is Dead In The AI Era

AI Accelerator Beachfront Considerations, Memory Pooling Negatives, Custom Silicon Adoption

Mar 16 •

and

Jeremie Eliahou Ontiveros

AI Datacenter Energy Dilemma - Race for AI Datacenter Space

Gigawatt Dreams and Matroyshka Brains Limited By Datacenters Not Chips

Mar 13 •

,

Daniel Nishball

, and

Jeremie Eliahou Ontiveros

February 2024

Groq Inference Tokenomics: Speed, But At What Cost?

Faster than Nvidia? Dissecting the economics

Feb 21 •

and

Daniel Nishball

Hybrid Bonding Process Flow - Advanced Packaging Part 5

BESI, EV Group, AMAT, TEL, ASMPT, SET, Shibaura, SUSS Microtec

Feb 9 •

,

, and

January 2024

Neural Network Quantization & Number Formats From First Principles

Inference & Training - Next Gen Hardware for Nvidia, AMD, Intel, Google, Microsoft, Meta, Arm, Qualcomm, MatX and Lemurian Labs

Jan 11 •

Intel GenAI For Yield, TSMC CFET & 3D Stacking, AMD 3D Device Modeling, Applied Materials Material Innovation, SK Hynix HBM4, Micron 3D DRAM…

China’s CXMT Sanctions Violation, Samsung 1000-layer Vertical NAND (VNAND), Kioxia highest density CMOS bonded to array (CBA) NAND, Micron non-volatile…

Jan 3 •

,

,

,

Daniel Nishball

, and

Anand Chamarthy

December 2023

Inference Race To The Bottom - Make It Up On Volume?

Mixtral Inference Costs on H100, MI300X, H200, A100, Speculative Decoding

Dec 18, 2023 •

and

Daniel Nishball

#nojs-banner { position: fixed; bottom: 0; left: 0; padding: 16px 16px 16px 32px; width: 100%; box-sizing: border-box; background: red; color: white; font-family: -apple-system, "Segoe UI", Roboto, Helvetica, Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol"; font-size: 13px; line-height: 13px; } #nojs-banner a { color: inherit; text-decoration: underline; } This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts