Industry Insights 14 min read

What’s Inside an AI Server? A Deep Dive into DGX A100 PCB Costs

This article dissects the NVIDIA DGX A100 AI server, breaking down its fan modules, storage, GPU board tray, CPU motherboard tray, power modules and other accessories to quantify PCB area usage and estimate the monetary value of each component, revealing how hardware choices drive overall server cost.

Architects' Tech Alliance
Architects' Tech Alliance
Architects' Tech Alliance
What’s Inside an AI Server? A Deep Dive into DGX A100 PCB Costs

Introduction

The analysis aims to reveal the internal composition of an AI server by examining the printed circuit board (PCB) components of the NVIDIA DGX A100, a benchmark product in the AI server market. By quantifying PCB area and assigning monetary values, the article helps readers understand where value is created during server upgrades.

Five Major Hardware Sections

The DGX A100 can be divided into five hardware sections: fan module, storage, GPU board tray, CPU motherboard tray, and power module.

1. Fan Module

The front fan module consists of eight fans, matching a typical 8U server configuration.

2. Storage

Eight 3.84 TB hard drives provide a total of 30 TB internal storage, positioned below the fan module.

3. GPU Board Tray

The GPU board tray is the core of an AI server and includes four PCB sub‑components: GPU carrier board, NVSwitch board, OAM (GPU accelerator card), and UBB (unit baseboard). The combined PCB area for the GPU tray is approximately 0.624 m², with a total estimated value of ¥12,250 per server (≈52 % carrier board, 48 % PCB).

GPU carrier board : 8 × FCBGA boards (70‑100 mm, 14‑16 layers) – ¥5,200 total.

NVSwitch : 6 × boards – ¥1,170 total.

OAM : 8 × cards, 0.03 m² each, using 20‑layer Ultra‑Low‑Loss CCL – ¥2,880 total.

UBB : 1 × board, 0.30 m², 26‑layer HDI – ¥3,000 total.

4. CPU Motherboard Tray

The CPU tray comprises CPU carrier boards, the main motherboard, and functional sub‑boards (memory, NIC, expansion cards). The total PCB area is about 0.662 m², with an estimated value of ¥2,845 per server (46 % carrier board, 40 % motherboard, 14 % functional boards).

CPU carrier board : 2 × boards – ¥1,300 total.

CPU motherboard : 0.38 m², 10‑12 layers, Low‑Loss CCL – ¥1,140 total.

Functional boards : memory cards, NICs, riser cards, OS driver board – ¥405 total.

5. Power Module and Other Accessories

The rear of the server houses six power supplies, each occupying ~0.019 m² of PCB. Additional accessories include the front control console board and the hard drive PCB (≈0.008 m² each). Using 6‑10 layer FR4/Mid‑Loss CCL material, the combined value of these accessories is estimated at ¥226 per server.

Total PCB Area and Value

Summing all sections, the DGX A100 uses about 1.474 m² of PCB material, resulting in a total estimated PCB‑related cost of ¥15,321 per server. The GPU board tray accounts for roughly 80 % of this cost (≈¥12,000), the CPU motherboard tray contributes 19 % (≈¥2,845), and other accessories make up the remaining 1 % (≈¥226). From a board‑level perspective, carrier‑board value is ¥7,670 (≈50 %) and pure PCB value is ¥7,651 (≈50 %).

Conclusion

The breakdown demonstrates that the majority of PCB‑related expense in an AI server resides in the GPU subsystem, highlighting where future cost‑optimization efforts should focus. Understanding the detailed PCB composition helps architects and buyers make informed decisions about server upgrades and component selection.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

AI serverPCB analysishardware costDGX A100GPU boardCPU motherboard
Architects' Tech Alliance
Written by

Architects' Tech Alliance

Sharing project experiences, insights into cutting-edge architectures, focusing on cloud computing, microservices, big data, hyper-convergence, storage, data protection, artificial intelligence, industry practices and solutions.

0 followers
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.