16.1c Direct Attach Copper (DAC) vs. Active Optical Cable (AOC) vs. pluggable optics¶
When building an AI data center, one of the first decisions you'll face is how to physically connect your switches, servers, and GPUs. The three main options are Direct Attach Copper (DAC) cables, Active Optical Cables (AOC), and pluggable optics (also called transceivers). Each has trade-offs in cost, power, distance, and reliability. This guide breaks down the differences in simple terms so you can choose the right cable for your AI cluster.
⚙️ What Are These Cables?¶
- Direct Attach Copper (DAC) — A copper cable with fixed connectors on both ends. It's like a thick, heavy HDMI cable but for data center networking. No separate transceiver module is needed.
- Active Optical Cable (AOC) — An optical fiber cable with built-in transceivers on both ends. It's lighter and thinner than copper but uses lasers to send data.
- Pluggable Optics — A separate transceiver module (like a QSFP or SFP) that plugs into a switch or server port. You then attach a separate fiber optic cable to the module.
📊 Key Comparison Table¶
| Feature | DAC | AOC | Pluggable Optics |
|---|---|---|---|
| Max Distance | Up to 7 meters | Up to 100 meters | Up to 10+ km (with fiber) |
| Power Usage | Very low (~0.1W) | Low (~1W) | Higher (~2-3.5W) |
| Cost | Lowest | Medium | Highest (module + cable) |
| Weight & Flexibility | Heavy, stiff | Light, flexible | Depends on cable type |
| Reliability | High (passive) | Good (active electronics) | Good (but more failure points) |
| Best For | Short rack-to-rack | Within a row or aisle | Long distances or future upgrades |
🛠️ When to Use Each Option¶
✅ DAC — The Budget Workhorse¶
- Use when: Connecting devices in the same rack or adjacent racks (under 5 meters).
- Why: Lowest cost, lowest power, and no extra modules to buy.
- Trade-off: Heavy and stiff — cable management is harder. Not suitable for long runs.
✅ AOC — The Flexible Middle Ground¶
- Use when: Connecting devices across a row or between rows (up to 100 meters).
- Why: Lighter than DAC, lower power than pluggable optics, and good for high-density setups.
- Trade-off: Slightly more expensive than DAC. The cable is fixed — you can't swap the ends.
✅ Pluggable Optics — The Long-Distance Specialist¶
- Use when: Connecting across data center halls, between buildings, or when you need future flexibility.
- Why: Can reach kilometers with single-mode fiber. You can upgrade the module without replacing the cable.
- Trade-off: Highest cost per link. More power consumption. Requires separate fiber cable and module.
📊 Visual Representation: DAC vs. AOC vs. Discrete Transceiver Cabling¶
This diagram contrasts Direct Attach Copper, Active Optical Cable, and discrete transceivers for different rack distance requirements.
🕵️ Real-World AI Deployment Tips¶
- For GPU-to-switch connections (within the same rack): Use DAC cables. They're cheap, low latency, and reliable.
- For switch-to-switch connections (across a row): Use AOC cables. They handle longer distances without signal loss.
- For top-of-rack to aggregation switches (across the data center): Use pluggable optics with single-mode fiber. This gives you the reach you need.
- Power budget: In a large AI cluster with thousands of links, choosing DAC over pluggable optics can save hundreds of watts of power.
- Cable management: DAC cables are heavy. Plan your cable trays and routing carefully to avoid airflow blockage.
📝 Quick Decision Flowchart¶
Question 1: Is the distance under 5 meters?
→ Yes: Use DAC
→ No: Go to Question 2
Question 2: Is the distance under 100 meters?
→ Yes: Use AOC
→ No: Use Pluggable Optics with fiber
🔍 Summary¶
- DAC = Cheap, short, low power — best for in-rack connections.
- AOC = Medium cost, medium distance, flexible — best for within-row connections.
- Pluggable Optics = Expensive, long distance, flexible — best for cross-hall or future-proofing.
As a new engineer, start with DAC for most GPU-to-switch links. Only move to AOC or pluggable optics when distance or future upgrade needs force you. This keeps your AI infrastructure simple, cost-effective, and easy to manage.