github-starred/cs249r_book
2
0
Fork 0
mirror of https://github.com/harvard-edge/cs249r_book.git synced 2026-03-11 17:49:25 -05:00
Code Issues Packages Projects Releases Wiki Activity

refactor: rename Vol I tagline from Operate to Deploy

Browse Source 
Update Volume I tagline from "Build, Optimize, Operate" to
"Build, Optimize, Deploy" across all documentation.

- "Deploy" better matches Part IV content (Serving, MLOps, Responsible Engineering)
- "Operate" implies ongoing management which is more Volume II territory
- Also fixes hw_acceleration table to use proper grid table format
This commit is contained in:
Vijay Janapa Reddi
2026-01-03 10:09:19 -05:00
parent 773897c07d
commit 95e0eafcdc
4 changed files with 25 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
book/README.md
View File

@@ -45,17 +45,17 @@ This textbook is organized into **two volumes** following the Hennessy & Patters
| Volume | Theme | Focus |
|--------|-------|-------|
| **Volume I** | Build, Optimize, Operate | Single-machine ML systems, foundational principles |
| **Volume I** | Build, Optimize, Deploy | Single-machine ML systems, foundational principles |
| **Volume II** | Scale, Distribute, Govern | Distributed systems at production scale |
#### Volume I: Build, Optimize, Operate
#### Volume I: Build, Optimize, Deploy
| Part | Focus | Chapters |
|------|-------|----------|
| **ML Foundations** | Core concepts | Introduction, ML Systems, DL Primer, Architectures |
| **System Development** | Building blocks | Workflow, Data Engineering, Frameworks, Training |
| **Model Optimization** | Making it fast | Efficient AI, Optimizations, HW Acceleration, Benchmarking |
| **System Operations** | Making it work | MLOps, Responsible Engineering |
| **Foundations** | Core concepts | Introduction, ML Systems, DL Primer, Architectures |
| **Development** | Building blocks | Workflow, Data Engineering, Frameworks, Training |
| **Optimization** | Making it fast | Efficient AI, Optimizations, HW Acceleration, Benchmarking |
| **Deployment** | Making it work | Serving, MLOps, Responsible Engineering |
#### Volume II: Scale, Distribute, Govern
@@ -124,7 +124,7 @@ cd book
book/
├── quarto/ # Book source (Quarto markdown)
│ ├── contents/ # Chapter content
│ │ ├── vol1/ # Volume I: Build, Optimize, Operate
│ │ ├── vol1/ # Volume I: Build, Optimize, Deploy
│ │ ├── vol2/ # Volume II: Scale, Distribute, Govern
│ │ ├── frontmatter/ # Preface, about, changelog
│ │ └── backmatter/ # References, glossary

4
book/docs/VOLUME_STRUCTURE.md
View File

@@ -10,14 +10,14 @@
This textbook is organized into two volumes following the Hennessy & Patterson pedagogical model:
- **Volume I: Build, Optimize, Operate** - Foundational knowledge for single-machine ML systems
- **Volume I: Build, Optimize, Deploy** - Foundational knowledge for single-machine ML systems
- **Volume II: Scale, Distribute, Govern** - Advanced distributed systems at production scale
Each volume stands alone as a complete learning experience while together forming a comprehensive treatment of the field.
---
## Volume I: Build, Optimize, Operate
## Volume I: Build, Optimize, Deploy
### Goal
A reader completes Volume I and can competently build, optimize, and deploy ML systems on a single machine with awareness of responsible practices.

6
book/quarto/contents/frontmatter/about/about.qmd
View File

@@ -18,12 +18,12 @@ This textbook is organized into two volumes following the **Hennessy & Patterson
| Volume | Theme | Focus | Analogy |
|--------|-------|-------|---------|
| **Volume I** | Build, Optimize, Operate | Single-machine ML systems, foundational principles | "Computer Organization and Design" |
| **Volume I** | Build, Optimize, Deploy | Single-machine ML systems, foundational principles | "Computer Organization and Design" |
| **Volume II** | Scale, Distribute, Govern | Distributed systems at production scale | "Computer Architecture" |
**Volume I** teaches you to *understand* ML systems. **Volume II** teaches you to *build* ML systems at scale.
**Volume I: Build, Optimize, Operate** establishes the foundations through four progressive stages:
**Volume I: Build, Optimize, Deploy** establishes the foundations through four progressive stages:
- **Foundations** (Part I): Build your conceptual foundation, establishing the mental models that underpin all effective systems work.
@@ -118,7 +118,7 @@ SocratiQ is still a work in progress, and we welcome your feedback to make it be
This work takes you from understanding ML systems conceptually to building and deploying them in practice. The content is organized into two volumes following the Hennessy & Patterson pedagogical model, each containing four parts that develop specific capabilities.
**Volume I: Build, Optimize, Operate**
**Volume I: Build, Optimize, Deploy**
1. **Part I: Foundations**
*Master the fundamentals.* Build intuition for how ML systems differ from traditional software, understand the hardware-software stack, and gain fluency with essential architectures and mathematical foundations.

20
book/quarto/contents/vol1/hw_acceleration/hw_acceleration.qmd
View File

@@ -1126,13 +1126,19 @@ The energy differential established earlier—where memory access costs dominate
@tbl-accelerator-economics provides concrete cost-performance data for representative accelerators, but the economic analysis must account for utilization efficiency and energy consumption patterns that determine real-world performance.
| **Accelerator** | **List Price (USD)** | **Peak FLOPS (FP16)** | **Memory Bandwidth** | **Price/Performance** |
|-----------------|----------------------|-----------------------|----------------------|-----------------------|
| NVIDIA V100 | ~$9,000 (2017-19) | 125 TFLOPS | 900 GB/s | $72/TFLOP |
| NVIDIA A100 | $15,000 | 312 TFLOPS (FP16) | 1,935 GB/s | $48/TFLOP |
| NVIDIA H100 | $25,000-30,000 | 756 TFLOPS (TF32) | 3,350 GB/s | $33/TFLOP |
| Google TPUv4 | ~$8,000* | 275 TFLOPS (BF16) | 1,200 GB/s | $29/TFLOP |
| Intel Gaudi 2 | $12,000 | 200 TFLOPS (INT8) | 800 GB/s | $60/TFLOP |
+-------------------+----------------------+-----------------------+----------------------+-----------------------+
| **Accelerator** | **List Price (USD)** | **Peak FLOPS (FP16)** | **Memory Bandwidth** | **Price/Performance** |
+==================:+=====================:+======================:+=====================:+======================:+
| **NVIDIA V100** | ~$9,000 (2017-19) | 125 TFLOPS | 900 GB/s | $72/TFLOP |
+-------------------+----------------------+-----------------------+----------------------+-----------------------+
| **NVIDIA A100** | $15,000 | 312 TFLOPS (FP16) | 1,935 GB/s | $48/TFLOP |
+-------------------+----------------------+-----------------------+----------------------+-----------------------+
| **NVIDIA H100** | $25,000-30,000 | 756 TFLOPS (TF32) | 3,350 GB/s | $33/TFLOP |
+-------------------+----------------------+-----------------------+----------------------+-----------------------+
| **Google TPUv4** | ~$8,000* | 275 TFLOPS (BF16) | 1,200 GB/s | $29/TFLOP |
+-------------------+----------------------+-----------------------+----------------------+-----------------------+
| **Intel Gaudi 2** | $12,000 | 200 TFLOPS (INT8) | 800 GB/s | $60/TFLOP |
+-------------------+----------------------+-----------------------+----------------------+-----------------------+
: **Accelerator Cost-Performance Comparison**: Hardware costs must be evaluated against computational capabilities to determine optimal deployment strategies. While newer accelerators like H100 offer better price-performance ratios, total cost of ownership includes power consumption, cooling requirements, and infrastructure costs that significantly impact operational economics. *TPU pricing estimated from cloud rates. {#tbl-accelerator-economics}