40G and 100G Optics: Which MPO or LC Cable Each One Needs
A transceiver that works and a link that is dead usually have the same cause: the cable was ordered against a drawing instead of against the optic. With optics now the long-lead item, the pairing has to be right the first time.
A transceiver that works and a link that is dead usually have the same cause: the cable was ordered against a drawing instead of against the optic. With optics now the long-lead item, the pairing has to be right the first time.
A transceiver that works and a link that is dead usually have the same cause: the cable was ordered against a drawing instead of against the optic. With optics now the long-lead item, the pairing has to be right the first time.
Optical is now growing faster than the chips it serves
Corning reported third quarter results yesterday. Core sales were $5.08 billion, up 19 percent. Optical Communications was $2.38 billion of that, up 44 percent. Enterprise Networks inside that segment grew 84 percent, and Corning's chief executive said Gen AI product sales doubled year over year, outpacing GPU unit growth.
That last comparison is the one worth sitting with. The optical layer is growing faster than the number of accelerators being installed, which only happens when each new rack needs more optical connections than the last one did.
At the same time the optics themselves have become the constrained part. Laser diode supply is the stated bottleneck across the transceiver industry, and manufacturers have been expanding fab capacity against order books that run years out.
Which puts a specific problem in front of anyone cabling a room. The transceiver is now the long-lead item and the cable has to match it exactly. Order the wrong cable and you have a working optic and a dead link. This is the guide to getting that pairing right the first time.
Corning reported third quarter results yesterday. Core sales were $5.08 billion, up 19 percent. Optical Communications was $2.38 billion of that, up 44 percent. Enterprise Networks inside that segment grew 84 percent, and Corning's chief executive said Gen AI product sales doubled year over year, outpacing GPU unit growth.
That last comparison is the one worth sitting with. The optical layer is growing faster than the number of accelerators being installed, which only happens when each new rack needs more optical connections than the last one did.
At the same time the optics themselves have become the constrained part. Laser diode supply is the stated bottleneck across the transceiver industry, and manufacturers have been expanding fab capacity against order books that run years out.
Which puts a specific problem in front of anyone cabling a room. The transceiver is now the long-lead item and the cable has to match it exactly. Order the wrong cable and you have a working optic and a dead link. This is the guide to getting that pairing right the first time.
Corning reported third quarter results yesterday. Core sales were $5.08 billion, up 19 percent. Optical Communications was $2.38 billion of that, up 44 percent. Enterprise Networks inside that segment grew 84 percent, and Corning's chief executive said Gen AI product sales doubled year over year, outpacing GPU unit growth.
That last comparison is the one worth sitting with. The optical layer is growing faster than the number of accelerators being installed, which only happens when each new rack needs more optical connections than the last one did.
At the same time the optics themselves have become the constrained part. Laser diode supply is the stated bottleneck across the transceiver industry, and manufacturers have been expanding fab capacity against order books that run years out.
Which puts a specific problem in front of anyone cabling a room. The transceiver is now the long-lead item and the cable has to match it exactly. Order the wrong cable and you have a working optic and a dead link. This is the guide to getting that pairing right the first time.
What the optic decides, and what it does not
A transceiver is a small module that converts electrical signals into light and back. Everything about the cable follows from how it does that.
The first division is parallel against duplex. A parallel optic splits the data across several fibers running side by side, each carrying one lane. A duplex optic puts everything on a single pair, one fiber transmitting and one receiving, using either a single wavelength or several wavelengths multiplexed together on the same strand.
That single choice sets the connector. Parallel optics need an MPO, because you cannot practically terminate eight or twenty fibers into anything smaller. Duplex optics need an LC pair. There is no adapter that makes one into the other, which is why a parallel optic and a duplex cable is such an expensive mistake.
The second division is multimode against singlemode. Multimode optics use cheap vertical cavity lasers at 850 nanometers and run short distances, typically inside a row or across a hall. Singlemode optics use more expensive lasers at 1310 or 1550 nanometers and run kilometers. The fiber in the cable has to match the optic, not the other way round, and the two are not interchangeable at any distance.
The third is reach. Every optic has a stated maximum distance on a given fiber grade, and that number is the budget for the entire channel including every connector and splice along the way, not just the cable length.
What the optic does not decide is the brand of anything. Two switches from different vendors running the same optic type take the same cable.
A transceiver is a small module that converts electrical signals into light and back. Everything about the cable follows from how it does that.
The first division is parallel against duplex. A parallel optic splits the data across several fibers running side by side, each carrying one lane. A duplex optic puts everything on a single pair, one fiber transmitting and one receiving, using either a single wavelength or several wavelengths multiplexed together on the same strand.
That single choice sets the connector. Parallel optics need an MPO, because you cannot practically terminate eight or twenty fibers into anything smaller. Duplex optics need an LC pair. There is no adapter that makes one into the other, which is why a parallel optic and a duplex cable is such an expensive mistake.
The second division is multimode against singlemode. Multimode optics use cheap vertical cavity lasers at 850 nanometers and run short distances, typically inside a row or across a hall. Singlemode optics use more expensive lasers at 1310 or 1550 nanometers and run kilometers. The fiber in the cable has to match the optic, not the other way round, and the two are not interchangeable at any distance.
The third is reach. Every optic has a stated maximum distance on a given fiber grade, and that number is the budget for the entire channel including every connector and splice along the way, not just the cable length.
What the optic does not decide is the brand of anything. Two switches from different vendors running the same optic type take the same cable.
A transceiver is a small module that converts electrical signals into light and back. Everything about the cable follows from how it does that.
The first division is parallel against duplex. A parallel optic splits the data across several fibers running side by side, each carrying one lane. A duplex optic puts everything on a single pair, one fiber transmitting and one receiving, using either a single wavelength or several wavelengths multiplexed together on the same strand.
That single choice sets the connector. Parallel optics need an MPO, because you cannot practically terminate eight or twenty fibers into anything smaller. Duplex optics need an LC pair. There is no adapter that makes one into the other, which is why a parallel optic and a duplex cable is such an expensive mistake.
The second division is multimode against singlemode. Multimode optics use cheap vertical cavity lasers at 850 nanometers and run short distances, typically inside a row or across a hall. Singlemode optics use more expensive lasers at 1310 or 1550 nanometers and run kilometers. The fiber in the cable has to match the optic, not the other way round, and the two are not interchangeable at any distance.
The third is reach. Every optic has a stated maximum distance on a given fiber grade, and that number is the budget for the entire channel including every connector and splice along the way, not just the cable length.
What the optic does not decide is the brand of anything. Two switches from different vendors running the same optic type take the same cable.
The options you will actually meet at 40G and 100G
At 40 gigabit the common parallel option is SR4, which uses four lanes at 10 gigabit each over multimode. It needs an MPO-12 connector, though only eight of the twelve fibers carry traffic, four transmitting and four receiving, with the middle four unused. Typical reach is around 100 meters on OM3 and 150 on OM4. An extended variant, eSR4, pushes that to roughly 300 and 400 meters respectively using the same connector and fiber.
Still at 40 gigabit, LR4 takes the duplex route. Four wavelengths around 1310 nanometers are multiplexed onto one singlemode pair, terminated LC, reaching 10 kilometers. PLR4 is the parallel singlemode variant, MPO-12 again but on singlemode, also to 10 kilometers.
At 100 gigabit the pattern repeats with more options. SR4 is four lanes at 25 gigabit over multimode on MPO-12, reaching roughly 70 meters on OM3 and 100 on OM4. The older SR10 implementation spreads ten lanes across an MPO-24 and is now rare in new builds.
On singlemode at 100 gigabit, PSM4 is parallel, MPO-12, 500 meters. CWDM4 is duplex, LC, 2 kilometers, using four wavelengths on one pair. LR4 is duplex, LC, 10 kilometers. ER4 reaches 40.
The practical reading is that at any given speed you are choosing between an MPO and multimode for short reach inside the room, an MPO and singlemode for mid-range parallel runs, or an LC pair and singlemode for anything long. The distance you need usually makes the decision for you, and the connector follows.
Above 100 gigabit the same logic continues into 200, 400 and beyond, with SR8 and DR4 style parallel options on MPO and FR4 and LR4 style duplex options on LC, and the fiber counts per module rising accordingly.
At 40 gigabit the common parallel option is SR4, which uses four lanes at 10 gigabit each over multimode. It needs an MPO-12 connector, though only eight of the twelve fibers carry traffic, four transmitting and four receiving, with the middle four unused. Typical reach is around 100 meters on OM3 and 150 on OM4. An extended variant, eSR4, pushes that to roughly 300 and 400 meters respectively using the same connector and fiber.
Still at 40 gigabit, LR4 takes the duplex route. Four wavelengths around 1310 nanometers are multiplexed onto one singlemode pair, terminated LC, reaching 10 kilometers. PLR4 is the parallel singlemode variant, MPO-12 again but on singlemode, also to 10 kilometers.
At 100 gigabit the pattern repeats with more options. SR4 is four lanes at 25 gigabit over multimode on MPO-12, reaching roughly 70 meters on OM3 and 100 on OM4. The older SR10 implementation spreads ten lanes across an MPO-24 and is now rare in new builds.
On singlemode at 100 gigabit, PSM4 is parallel, MPO-12, 500 meters. CWDM4 is duplex, LC, 2 kilometers, using four wavelengths on one pair. LR4 is duplex, LC, 10 kilometers. ER4 reaches 40.
The practical reading is that at any given speed you are choosing between an MPO and multimode for short reach inside the room, an MPO and singlemode for mid-range parallel runs, or an LC pair and singlemode for anything long. The distance you need usually makes the decision for you, and the connector follows.
Above 100 gigabit the same logic continues into 200, 400 and beyond, with SR8 and DR4 style parallel options on MPO and FR4 and LR4 style duplex options on LC, and the fiber counts per module rising accordingly.
At 40 gigabit the common parallel option is SR4, which uses four lanes at 10 gigabit each over multimode. It needs an MPO-12 connector, though only eight of the twelve fibers carry traffic, four transmitting and four receiving, with the middle four unused. Typical reach is around 100 meters on OM3 and 150 on OM4. An extended variant, eSR4, pushes that to roughly 300 and 400 meters respectively using the same connector and fiber.
Still at 40 gigabit, LR4 takes the duplex route. Four wavelengths around 1310 nanometers are multiplexed onto one singlemode pair, terminated LC, reaching 10 kilometers. PLR4 is the parallel singlemode variant, MPO-12 again but on singlemode, also to 10 kilometers.
At 100 gigabit the pattern repeats with more options. SR4 is four lanes at 25 gigabit over multimode on MPO-12, reaching roughly 70 meters on OM3 and 100 on OM4. The older SR10 implementation spreads ten lanes across an MPO-24 and is now rare in new builds.
On singlemode at 100 gigabit, PSM4 is parallel, MPO-12, 500 meters. CWDM4 is duplex, LC, 2 kilometers, using four wavelengths on one pair. LR4 is duplex, LC, 10 kilometers. ER4 reaches 40.
The practical reading is that at any given speed you are choosing between an MPO and multimode for short reach inside the room, an MPO and singlemode for mid-range parallel runs, or an LC pair and singlemode for anything long. The distance you need usually makes the decision for you, and the connector follows.
Above 100 gigabit the same logic continues into 200, 400 and beyond, with SR8 and DR4 style parallel options on MPO and FR4 and LR4 style duplex options on LC, and the fiber counts per module rising accordingly.
Breakout, direct attach, and the mistakes that cost a week
Parallel optics make breakout possible, and it is often the cheapest capacity in a room. A single 40 gigabit SR4 can be split into four independent 10 gigabit links, and a 100 gigabit SR4 into four 25 gigabit links, using a harness that takes the MPO at one end and presents four duplex LC pairs at the other. One switch port becomes four. The constraint is that the switch has to support breakout mode, and the ports at the far end have to be the matching speed.
For very short runs there are two alternatives to optics entirely. A direct attach copper cable, or DAC, connects two ports with copper and no transceiver at all, which is cheap and draws almost no power but is limited to a few meters and is bulky. An active optical cable, or AOC, has the optics permanently attached at both ends, which removes the connector mismatch problem but also means the cable and the optics fail together.
The mistakes are consistent and expensive. Ordering a duplex cable for a parallel optic, or the reverse. Buying multimode cable for singlemode optics because the price looked better. Specifying reach from the drawing rather than the routed cable length plus every connector in the path. Assuming SR4 at 40 gigabit and SR4 at 100 gigabit take the same cable, which they do for the connector but not for the distance you will get. And treating breakout as free when the switch does not support it.
Since the optics are now the long-lead item, the order that matters is to settle the transceiver first and let it specify the cable, rather than ordering cable against a drawing and discovering the mismatch when the modules arrive.
Parallel optics make breakout possible, and it is often the cheapest capacity in a room. A single 40 gigabit SR4 can be split into four independent 10 gigabit links, and a 100 gigabit SR4 into four 25 gigabit links, using a harness that takes the MPO at one end and presents four duplex LC pairs at the other. One switch port becomes four. The constraint is that the switch has to support breakout mode, and the ports at the far end have to be the matching speed.
For very short runs there are two alternatives to optics entirely. A direct attach copper cable, or DAC, connects two ports with copper and no transceiver at all, which is cheap and draws almost no power but is limited to a few meters and is bulky. An active optical cable, or AOC, has the optics permanently attached at both ends, which removes the connector mismatch problem but also means the cable and the optics fail together.
The mistakes are consistent and expensive. Ordering a duplex cable for a parallel optic, or the reverse. Buying multimode cable for singlemode optics because the price looked better. Specifying reach from the drawing rather than the routed cable length plus every connector in the path. Assuming SR4 at 40 gigabit and SR4 at 100 gigabit take the same cable, which they do for the connector but not for the distance you will get. And treating breakout as free when the switch does not support it.
Since the optics are now the long-lead item, the order that matters is to settle the transceiver first and let it specify the cable, rather than ordering cable against a drawing and discovering the mismatch when the modules arrive.
Parallel optics make breakout possible, and it is often the cheapest capacity in a room. A single 40 gigabit SR4 can be split into four independent 10 gigabit links, and a 100 gigabit SR4 into four 25 gigabit links, using a harness that takes the MPO at one end and presents four duplex LC pairs at the other. One switch port becomes four. The constraint is that the switch has to support breakout mode, and the ports at the far end have to be the matching speed.
For very short runs there are two alternatives to optics entirely. A direct attach copper cable, or DAC, connects two ports with copper and no transceiver at all, which is cheap and draws almost no power but is limited to a few meters and is bulky. An active optical cable, or AOC, has the optics permanently attached at both ends, which removes the connector mismatch problem but also means the cable and the optics fail together.
The mistakes are consistent and expensive. Ordering a duplex cable for a parallel optic, or the reverse. Buying multimode cable for singlemode optics because the price looked better. Specifying reach from the drawing rather than the routed cable length plus every connector in the path. Assuming SR4 at 40 gigabit and SR4 at 100 gigabit take the same cable, which they do for the connector but not for the distance you will get. And treating breakout as free when the switch does not support it.
Since the optics are now the long-lead item, the order that matters is to settle the transceiver first and let it specify the cable, rather than ordering cable against a drawing and discovering the mismatch when the modules arrive.
Getting both halves from the same conversation
Getting both halves from the same conversation
Getting both halves from the same conversation
The pairing problem exists because optics and cable are usually bought from different suppliers, on different lead times, against different drawings. Nobody owns the join.
Vocom International supplies both sides of it rather than manufacturing either. The range covers high performance transceivers and active optical cable alongside factory terminated MPO and MTP trunks, harness breakouts, cassettes and LC and SC patch, and the high count outside plant cable that feeds the room, all sourced build-to-order through tier 1 manufacturing partners. Standard lead time is 6 to 8 weeks, with 2 to 3 week expedited air freight available and up to 10 weeks on custom builds. For the trunk side of the decision, our guide to MPO and MTP trunks covers polarity and breakout in detail.
If you have a module list and a cable schedule that were written by different people, that is the pairing worth checking before either one is ordered. Talk to a specifier.
The pairing problem exists because optics and cable are usually bought from different suppliers, on different lead times, against different drawings. Nobody owns the join.
Vocom International supplies both sides of it rather than manufacturing either. The range covers high performance transceivers and active optical cable alongside factory terminated MPO and MTP trunks, harness breakouts, cassettes and LC and SC patch, and the high count outside plant cable that feeds the room, all sourced build-to-order through tier 1 manufacturing partners. Standard lead time is 6 to 8 weeks, with 2 to 3 week expedited air freight available and up to 10 weeks on custom builds. For the trunk side of the decision, our guide to MPO and MTP trunks covers polarity and breakout in detail.
If you have a module list and a cable schedule that were written by different people, that is the pairing worth checking before either one is ordered. Talk to a specifier.
The pairing problem exists because optics and cable are usually bought from different suppliers, on different lead times, against different drawings. Nobody owns the join.
Vocom International supplies both sides of it rather than manufacturing either. The range covers high performance transceivers and active optical cable alongside factory terminated MPO and MTP trunks, harness breakouts, cassettes and LC and SC patch, and the high count outside plant cable that feeds the room, all sourced build-to-order through tier 1 manufacturing partners. Standard lead time is 6 to 8 weeks, with 2 to 3 week expedited air freight available and up to 10 weeks on custom builds. For the trunk side of the decision, our guide to MPO and MTP trunks covers polarity and breakout in detail.
If you have a module list and a cable schedule that were written by different people, that is the pairing worth checking before either one is ordered. Talk to a specifier.
