© 2025–2026 Subramaniyam (Sam) Pooni
All Rights Reserved
Proprietary & Confidential
Visual Reference

Figure Gallery

Every figure in the package, listed from the filesystem. 70 live figures across 14 chapters, 12 of them new in v6.0, plus a 36-file v4.0 archive retained only so the retraction stays checkable.

70
Live Figures
12
New in v6.0
14
Chapters
36
Superseded (v4.0)
This page is generated from the filesystem. Every card below is a file that exists in figures/; every link was resolved against disk when the page was built. Counts are not hand-maintained. “Embedded in” is read from the <iframe> tags in chapters/, so a figure marked gallery only is genuinely not embedded in any chapter body.

CH00 — Executive Summary 7 figures

Figure 2.1 — LLM Inference Phases
figure-2-1-inference-phases.html
Gallery only — not embedded in a chapter body
Figure 2.2 — KV-Cache Growth
figure-2-2-kv-cache-growth.html
Gallery only — not embedded in a chapter body
Figure 7.2 — Attention Locality Patterns
figure-3-3-attention-locality.html
Gallery only — not embedded in a chapter body
Figure 7.3 — GQA Structure
figure-3-4-gqa-structure.html
Gallery only — not embedded in a chapter body
Figure 7.9 — EMA Eviction Policy
figure-4-2-ema-eviction.html
Gallery only — not embedded in a chapter body
KV-Cache Offloading: Competitive Landscape
competitive-landscape.html
Gallery only — not embedded in a chapter body
Visual appendix
visual-appendix.html
Gallery only — not embedded in a chapter body

CH01 — Introduction 3 figures

Distributed endpoint diagrams
distributed-endpoint-diagrams.html
Gallery only — not embedded in a chapter body
KV-Cache Math Explainer
kv-cache-math-explainer.html
Gallery only — not embedded in a chapter body
Visual appendix
visual-appendix.html
Gallery only — not embedded in a chapter body

CH02 — Background 9 figures

Bandwidth Utilisation
bandwidth-utilisation.html
Figure 2.7 · Measured
Gallery only — not embedded in a chapter body
Current Approaches Comparison
current-approaches-comparison.html
Gallery only — not embedded in a chapter body
Figure 2.1 — LLM Inference Phases
figure-2-1-inference-phases.html
Gallery only — not embedded in a chapter body
Figure 2.2 — KV-Cache Growth
figure-2-2-kv-cache-growth.html
Gallery only — not embedded in a chapter body
Figure 2.3 — Why KV-Cache
figure-2-3-why-kv-cache.html
Gallery only — not embedded in a chapter body
Figure 2.4 — Roofline Model
figure-2-4-roofline.html
Gallery only — not embedded in a chapter body
Section 2.1 — Inference Fundamentals
section-2-1-inference-fundamentals.html
Gallery only — not embedded in a chapter body
Two-Term Decode Model
measured-two-term-decode.html
Figure 2.6 · Measured
Gallery only — not embedded in a chapter body
Visual appendix
visual-appendix.html
Gallery only — not embedded in a chapter body

CH03 — The KV-State Control Plane 8 figures · 2 new

New · v6.0
KV-State Control Plane
kv-state-control-plane.html
Figure 3.2 · Illustrative — proposed architecture · one stage Simulated (provisional)
Embedded in Chapter 3
New · v6.0
Two Orthogonal Hierarchies
two-hierarchies.html
Figure 3.1 · Illustrative — proposed architecture
Embedded in Chapter 3
CXL Architecture Diagram
cxl-architecture-diagram.html
Gallery only — not embedded in a chapter body
CXL Architecture Diagram
figure-3-1-cxl-architecture.html
Gallery only — not embedded in a chapter body
Distributed Endpoint Architecture
diagram-architecture-html.html
Gallery only — not embedded in a chapter body
Endpoint diagram
endpoint-diagram.html
Gallery only — not embedded in a chapter body
Figure 3.2 — Memory Hierarchy
figure-3-2-memory-hierarchy.html
Gallery only — not embedded in a chapter body
Four Constraints
four-constraints.html
Figure 3.1 · Analytical model
Gallery only — not embedded in a chapter body

CH04 — Hardware Measurement 5 figures · 1 new

New · v6.0
Movement vs Regeneration
movement-vs-regeneration.html
Figure 5.1 · Analytical model
Embedded in Chapter 5
CXL.mem vs PCIe Latency Paths
diagram-latency-path-html.html
Gallery only — not embedded in a chapter body
Effective Latency Analysis
effective-latency-analysis.html
Gallery only — not embedded in a chapter body
Figure 4.3 — Latency Waterfall
figure-2-5-latency-waterfall.html
Gallery only — not embedded in a chapter body
Latency Calculations
latency-calculations.html
Gallery only — not embedded in a chapter body

CH05 — Bandwidth and Tier Economics 2 figures · 1 new

New · v6.0
Sustainability vs Hiding
sustainability-vs-hiding.html
Figure 5.2 · Analytical model
Embedded in Chapter 5
Bandwidth Aggregation
bandwidth-aggregation.html
Gallery only — not embedded in a chapter body

CH06 — Preprocessing Offload 1 figure

Preprocessing Offload
preprocessing-offload.html
Gallery only — not embedded in a chapter body

CH07 — Residency and Policy State 19 figures · 4 new

New · v6.0
Erratum: _pick_victim()
erratum-pick-victim.html
Figure 7.3 · Simulated (provisional)
Embedded in Chapter 7
New · v6.0
KV Block Object
kv-block-object.html
Figure 7.1 · Illustrative — object model specification
Embedded in Chapter 7
New · v6.0
Lossless vs Lossy Boundary
lossless-vs-lossy.html
Figure 7.2 · Analytical model
Embedded in Chapter 7
New · v6.0
Movement Amplification
movement-amplification.html
Figure 7.4 · Analytical model
Embedded in Chapter 7
Attention Patterns & RoPE Locality
diagram-memory-wall-html.html
Gallery only — not embedded in a chapter body
Attention-Aware Block Eviction (LRFU)
ema-eviction.html
Gallery only — not embedded in a chapter body
Figure 7.10 — RoPE Distance Bound
figure-4-3-rope-prefetch.html
Gallery only — not embedded in a chapter body
Figure 7.2 — Attention Locality Patterns
figure-3-3-attention-locality.html
Gallery only — not embedded in a chapter body
Figure 7.3 — GQA Structure
figure-3-4-gqa-structure.html
Gallery only — not embedded in a chapter body
Figure 7.3 — GQA Structure
figure-7-3-gqa-structure.html
Gallery only — not embedded in a chapter body
Figure 7.9 — LRFU Block Eviction
figure-4-2-ema-eviction.html
Gallery only — not embedded in a chapter body
GQA Per-Head Scoring Streams
gqa-tracking-diagram.html
Gallery only — not embedded in a chapter body
Grouped Query Attention
gqa-explainer.html
Gallery only — not embedded in a chapter body
LRFU Decayed-Frequency Scoring
diagram-ema-scoring-html.html
Gallery only — not embedded in a chapter body
LRFU Score and Eviction Priority
ema-calculation.html
Gallery only — not embedded in a chapter body
Per-Head Scoring Granularity
diagram-per-head-html.html
Gallery only — not embedded in a chapter body
Per-Head Scoring Signals
per-head-tracking.html
Gallery only — not embedded in a chapter body
RoPE Geometry Explainer
rope-prefetch.html
Gallery only — not embedded in a chapter body
RoPE Geometry — Worked Example
rope-prefetch-example.html
Gallery only — not embedded in a chapter body

CH08 — MoE Routing 1 figure

MoE Routing Support
moe-routing.html
Gallery only — not embedded in a chapter body

CH09 — GPU Integration 2 figures

GPU Integration
gpu-integration.html
Gallery only — not embedded in a chapter body
GPU–Tier KV-Cache Interaction
gpu-cxl-kvcache.html
Gallery only — not embedded in a chapter body

CH10 — The 2026 Landscape 4 figures · 1 new

New · v6.0
2026 Landscape
landscape-2026.html
Figure 10.1 · External literature
Embedded in Chapter 10
2026 Landscape by Decision
market-landscape.html
Gallery only — not embedded in a chapter body
The 2026 KV-Cache Landscape
competitive-landscape.html
Gallery only — not embedded in a chapter body
The 2026 KV-Cache Map
kv-cache-landscape.html
Gallery only — not embedded in a chapter body

CH11 — Results Summary 5 figures · 2 new

New · v6.0
Alpha Grid
alpha-grid.html
Figure 11.1 · Simulated (provisional)
Embedded in Chapter 11
New · v6.0
Persistent Metadata
persistent-metadata.html
Figure 11.2 · Simulated (provisional)
Embedded in Chapter 11
Figure 11.1 — The KV-State Control Plane
figure-4-1-full-architecture.html
Gallery only — not embedded in a chapter body
Figure 11.2 — Residency-State Persistence: The Largest Effect in the Study
figure-5-1-performance-benchmarks.html
Gallery only — not embedded in a chapter body
Policy Sweep
policy-sweep.html
Figure 11.1 · Simulated (provisional)
Gallery only — not embedded in a chapter body

CH12 — A Decision Procedure 3 figures · 1 new

New · v6.0
Decision Procedure
decision-procedure.html
Figure 12.1 · Illustrative — decision procedure
Embedded in Chapter 12
Cache Management Placement
cache-management-placement.html
Gallery only — not embedded in a chapter body
KV Cache Offloading — Visual Diagrams (v6.0)
kv-cache-diagrams-html.html
Gallery only — not embedded in a chapter body

CH13 — Conclusion 1 figure

Distributed Endpoint Architecture
cover-page.html
Gallery only — not embedded in a chapter body

Superseded — v4.0 archive (36 files)

These figures are retracted. They are retained, unmodified, for one reason only: so the retraction stays checkable. Every headline number they carry — the 97% hit rate, the 72→80→87→93→97 ladder, +25 points over LRU, 16× user capacity, 65×, 11.5×, 15.6×, 36% CapEx — is listed as retracted in RETRACTIONS.md and §7 of CANONICAL-NUMBERS.md. Nothing here was dishonest; all of it was constructed rather than observed, and the construction did not survive contact with a measurement. Do not cite anything on this shelf. Use the live figures above.
Bandwidth Aggregation
bandwidth-aggregation.html
Cache Management Placement
cache-management-placement.html
KV-Cache Offloading: Competitive Landscape
competitive-landscape.html
Distributed Endpoint Architecture
cover-page.html
Current Approaches Comparison
current-approaches-comparison.html
Distributed Endpoint Architecture
diagram-architecture-html.html
EMA Attention Scoring
diagram-ema-scoring-html.html
CXL.mem vs PCIe Latency Paths
diagram-latency-path-html.html
Attention Patterns & RoPE Locality
diagram-memory-wall-html.html
Per-Head Attention Tracking
diagram-per-head-html.html
Distributed endpoint diagrams
distributed-endpoint-diagrams.html
Effective Latency Analysis
effective-latency-analysis.html
EMA Calculation
ema-calculation.html
Attention-Aware Eviction
ema-eviction.html
Figure 11.1 — Full Architecture
figure-11-1-full-architecture.html
Figure 11.2 — Modeled System Comparison
figure-11-2-performance-benchmarks.html
Figure 2.4 — Roofline Model
figure-2-4-roofline.html
CXL Architecture Diagram
figure-3-1-cxl-architecture.html
Figure 3.2 — Memory Hierarchy
figure-3-2-memory-hierarchy.html
Figure 4.3 — Latency Waterfall
figure-4-3-latency-waterfall.html
Figure 7.10 — RoPE Prefetch
figure-7-10-rope-prefetch.html
Figure 7.2 — Attention Locality Patterns
figure-7-2-attention-locality.html
Figure 7.9 — EMA Eviction Policy
figure-7-9-ema-eviction-policy.html
Gpu cxl kvcache
gpu-cxl-kvcache.html
GPU Integration
gpu-integration.html
Grouped Query Attention
gqa-explainer.html
GQA Per-Head Tracking
gqa-tracking-diagram.html
KV Cache Offloading - Visual Diagrams & Architecture
kv-cache-diagrams-html.html
KV Cache Offloading: Market Landscape
kv-cache-landscape.html
Latency Calculations
latency-calculations.html
Market Landscape
market-landscape.html
Per-Head Tracking Rationale
per-head-tracking.html
Preprocessing Offload
preprocessing-offload.html
RoPE Prefetch - Worked Example
rope-prefetch-example.html
RoPE Prefetch Strategy
rope-prefetch.html
Visual appendix
visual-appendix.html