© 2025 Subramaniyam (Sam) Pooni
All Rights Reserved
Proprietary & Confidential

Visual Reference

Figure Gallery

114 diagrams across 51 source files covering all aspects of the KV-Cache offloading architecture.

114

Total Diagrams

51

Source Files

14

Chapters

27

KV-Cache Figs

CH00 — Executive Summary 14 diagrams / 6 files

Visual Appendix (30 diagrams)

Inference Phases

KV-Cache Growth

Attention Locality

CH01 — Introduction 5 diagrams / 3 files

Visual Appendix

Distributed Endpoint Diagrams

KV-Cache Math Explainer

CH02 — Background 8 diagrams / 7 files

Visual Appendix

Inference Phases

KV-Cache Growth

Current Approaches

Inference Fundamentals

CH03 — Architecture 11 diagrams / 5 files

Architecture Diagram

Endpoint Diagram

CXL Architecture

CXL Architecture Detail

Memory Hierarchy

CH04 — Latency 9 diagrams / 4 files

Effective Latency Analysis

Latency Waterfall

Latency Calculations

CH05 — Bandwidth 5 diagrams / 1 file

Bandwidth Aggregation

CH06 — Preprocessing 3 diagrams / 1 file

Preprocessing Offload

CH07 — KV-Cache Management 27 diagrams / 14 files

Per-Head Tracking

EMA Calculation

Per-Head Tracking Detail

RoPE Prefetch Example

Attention Locality

EMA Eviction Component

RoPE Prefetch Component

CH08 — MoE Routing 5 diagrams / 1 file

CH09 — GPU Integration 7 diagrams / 2 files

GPU Integration

GPU-CXL KV-Cache

CH10 — Market Landscape 6 diagrams / 2 files

Market Landscape

KV-Cache Landscape

CH11 — Performance 6 diagrams / 2 files

Full Architecture

Performance Benchmarks

CH12 — Implementation 5 diagrams / 2 files

Cache Management Placement

KV-Cache Diagrams

CH13 — Conclusion 3 diagrams / 1 file

← Back toHome Start Reading →Executive Summary