800G Transceiver Selection Guide: OSFP vs QSFP-DD800, DR8 vs 2×FR4, and Power Classes Explained
How to choose 800G optics in 2026: form factor decision (OSFP vs QSFP-DD800), PMD selection by reach (VR8/DR8/SR8/2×FR4), power classes and thermal planning, breakout compatibility with 400G, and a checklist that avoids the common ordering mistakes.
Conclusion first
800G selection is decided by three questions in order: What cage does my switch have? What reach does my cabling plan have? How much power can my chassis cool? Get those three right and the rest — SR8/DR8/2×FR4, breakout, vendors — falls into place. This guide walks each question with 2026-specific numbers.
Browse real 400G/800G module part numbers at Canjintech
>1. Form factor: OSFP vs QSFP-DD800| Attribute | OSFP | QSFP-DD800 |
|---|---|---|
| Footprint | Larger (about 20% wider body) | Same as QSFP-DD (dense 4× per 1U) |
| Thermal design | Integral heatsink options; best heat spreading | Cage-dependent; relies on airflow over module |
| Typical power envelope | Up to ~20 W classes comfortable | 16 W mainstream; higher classes need verified airflow |
| Port density per 1U | 32-40 typical | 48-64 typical |
| AI cluster presence | Dominant in 800G AI switches | Common in cloud/general 800G |
Decision rule: if you run AI/ML leaf-spine with high ambient temperatures or future 1.6T plans, OSFP buys thermal headroom. If you maximize port density in a conventional datacenter, QSFP-DD800 wins. Mixing both means maintaining two optics inventories — avoid it unless you genuinely have both platforms.
2. Reach and PMD selection
| PMD | Media | Reach class | Lanes | Typical use |
|---|---|---|---|---|
| 800GBASE-VR8 | VCSEL MMF | ~50 m | 8 × 100G | Same-row, low-cost top-of-rack |
| 800GBASE-SR8 | VCSEL MMF | ~100 m | 8 × 100G | In-row and adjacent-row |
| 800GBASE-DR8 | SMF (MPO) | ~500 m | 8 × 100G | Row-to-row, structured cabling |
| 2 × 400GBASE-DR4/FR4 | SMF (duplex) | 500 m / 2 km | 2 groups of 4 | Duplex-LC legacy plants |
| 800GBASE-2×FR4 | SMF (duplex) | ~2 km | 2 × 4 × 100G CWDM | Campus / inter-building |
| 800ZR (OIF coherent) | SMF (duplex) | 80-120 km | Coherent | DCI, not intra-DC |
Selection heuristics:
- New structured cabling with MPO trunks → DR8. Simplest inventory: one SKU, one trunk type.
- Existing duplex-LC plant (CWDM4/LR4 lineage) → 2×FR4 or 2×DR4 via breakout, avoids re-cabling.
- Same-rack and power-capped → consider VR8/SR8 or validated LPO SKUs.
- Anything beyond 2 km inside the metro → price a coherent 800ZR pair before assuming LR-class PAM4 exists (it mostly doesn't at 800G).
3. Power classes and thermal planning
| Power class (CMIS) | Approx. power | What you must verify |
|---|---|---|
| Class 5-6 | ≤ 8 W | Legacy QSFP-DD cages, check MSA compliance |
| Class 7-8 | 8-14 W | Common for 800G FR/LR classes |
| High-power classes | 14-20 W | Cage airflow spec + switch vendor qualified list |
Per-rack arithmetic worth doing before purchase:
Optics power = ports × per-module power
64-port × 16 W = 1024 W per switch face (800G)Add cage airflow derating (front-to-back vs side-to-air), and confirm the switch datasheet lists the specific module power class as supported — not just the form factor.
4. Breakout and backward compatibility
- 800G → 2 × 400G: most 800G DR8 modules have a 2×400G-DR4 breakout sibling; verify your switch supports per-port lane splitting in software (CMIS).
- 800G cage hosting 400G modules: supported on most OSFP/QSFP-DD800 platforms at reduced rate — a useful bridge strategy while your spine layer catches up.
- 400G QSFP-DD in QSFP-DD800 cages: mechanically and electrically compatible; this is the one direction of mixing that "just works" broadly.
- OSFP ↔ QSFP-DD: no cross-compatibility, full stop.
5. Ordering checklist
- Switch model → native cage and qualified module list.
- Cabling plan → reach and connector type (MPO vs duplex LC).
- Power class → matches cage airflow and switch qualified list.
- CMIS version → 4.0+ for scriptable DDM and lane mapping.
- Breakout plan → does your automation expect 1×800G or 2×400G ports?
- Spares → same SKU family as production; mixing vendors inside one rail group complicates DDM baselines.
FAQ
The FAQ block above answers the three most common ordering questions (OSFP vs QSFP-DD800, 400G in 800G ports, power budgeting).
Selection data integrated from QSFP-DD800 and OSFP MSA specifications, CMIS power class tables, and switch vendor qualification matrices — restructured with engineering commentary, no vendor text reproduced.
FAQ
- Are OSFP and QSFP-DD800 interchangeable?
- No. Different cage, connector and mechanical design. QSFP-DD800 reuses the QSFP-DD footprint (5 W-class legacy, up to 16 W with DD800 thermal extensions); OSFP is physically larger with better heat spreading. Your switch's native cage decides — there is no passive adapter.
- Can an 800G port run 400G modules?
- Yes, on most platforms: an 800G OSFP/QSFP-DD800 port can run 400G modules (DR4/2×FR4) at reduced rate, and many 800G modules support 2×400G breakout via front-panel ports. Confirm CMIS lane mapping support in your switch software.
- What power budget should I plan per 800G port?
- Design for 16 W per module as the 2026 planning number (some classes reach 18-20 W worst case). A 64-port switch then needs roughly 1 kW of optics headroom in cooling, independent of the ASIC.