Enter for the full results page · Esc to close
Transceivers are now a serious fraction of a switch's power draw, and all of it turns into heat inside the rack. Enter the port build and this gives you the module load, the switch total, the heat in BTU/hr, and how much of the rack's power budget is left.
Watts in, heat out. Essentially all the electrical power a transceiver draws leaves as heat inside the switch chassis. One watt is 3.412 BTU/hr, and twelve thousand BTU/hr is one ton of refrigeration. A rack of four 32-port 800G switches is dissipating well over ten kilowatts, which is about three tons of cooling for one rack.
Optics have become the dominant term. It used to be safe to size a rack on switch power and ignore the optics. At 1.5 W a port that was fine. At 16 W a port with 32 ports populated, the optics alone draw 512 W \u2014 comparable to the switch itself. Any rack power model that still treats transceivers as a rounding error will be wrong by a wide margin at 800G and hopeless at 1.6T.
The cheapest watts are the ones you do not use. Three levers, in order of how much they save. Use DAC or ACC for anything inside the rack \u2014 a passive copper link is a fraction of a watt against 32 W for a pair of optical modules. Pick the lowest-reach variant that clears the distance: DR4 over SR8 saves around 3 W per port, and choosing an ER-class part for a 200 m link wastes power at both ends. And check the actual populated port count \u2014 half-populating two switches costs more than fully populating one.
Faceplate temperature, not just total watts. A rack can be inside its kilowatt budget and still have modules running hot, because the constraint at the module is airflow across the cage. High-power optics in the middle of a dense faceplate run hotter than the same part at the end of the row. If DDM is reporting temperatures near the upper limit, the problem is airflow, not the power budget.
They are typical maximums for each class and good for planning within about 15%. Real parts vary with vendor, reach variant and how hard the DSP is working. For a final design, take the maximum figure from the datasheet of the exact part \u2014 the difference between a typical and a worst-case 800G module can be 3 W a port, which is 100 W across a switch.
Plan the power feed and cooling on maximum, and estimate the energy bill on typical. Sizing a breaker on typical draw is how you discover the problem during a heatwave rather than during design.
Yes, and it is the single biggest lever inside a rack. A passive DAC is a piece of copper: roughly 0.3 W against 32 W for the pair of optical modules it replaces. Over 32 in-rack links that is a kilowatt saved in one switch. The trade-off is reach and bulk \u2014 see our DAC and AOC range for where the limits fall.
Three-year cost of ownership per port — module price, power, cooling and failures
Open tool →Turn a switch port count and a breakout ratio into the exact optics and cables to order
Open tool →Pick a standard and get the wavelength plan, fibre type, connector, reach and planning loss budget
Open tool →Transceivers, MPO assemblies, WDM and patch cords — built and tested to the figures you just calculated.