Two lines, two limits, and the one that stops you first

That is not an inference from a datasheet footnote — it falls out of the published figures. NovaStar quote three per-port capacities for the MX40 Pro's gigabit ports at 60 Hz. With power-of-two containers, one link efficiency of 0.95 reproduces all three exactly, and the build below fails if any of them misses by a single pixel.

• The published ratios are the container table. 1.00, 0.75 and 0.50 are exactly 24/24, 24/32 and 24/48. NovaStar do not explain the packing anywhere; they simply print three numbers whose ratios can only be those containers.
• No single efficiency rescues the naive formula. Fit one to each depth and you need 0.950 at 8-bit, 0.891 at 10-bit and 0.712 at 12-bit — a spread of 33 %. A physical link does not change its framing overhead because the colour got deeper.
• Efficiency is not waste you can tune away. The missing 5 % is Ethernet framing, preamble, inter-packet gap and the vendor's own protocol. Brompton's Tessera links pack naively and run nearer 0.756; the constant belongs to the controller, not to the cable.

A panel's datasheet advertises 3840 Hz or 7680 Hz refresh, and it is tempting to put that number in the formula. It does not belong there. The frame crosses the wire once per video frame. The receiving card writes it to local memory and the driver ICs re-scan it from there, thousands of times a second, to build greyscale by pulse-width modulation. None of that rescanning touches the link.

Frame rate is the one that costs bandwidth, and it costs it proportionally: 120 Hz halves the port. The same confusion runs the other way round with bit depth — the panel's internal greyscale depth (14 to 16 bit, typically) is not the 8, 10 or 12-bit depth of the video on the wire, and feeding a panel deeper colour does not change the refresh rate it can achieve.

Absen's PL V2 user manual is the only manufacturer document found that tabulates this directly: the quantity of cabinets one 3×2.5 mm² power cable carries, at AC220 V and at AC110 V, for eighteen models. Two things fall out of checking it, and neither is stated in the manual.

• It is a current limit, not a power one. The 110 V column is the 220 V column halved — ⌈n/2⌉ in all eighteen rows. Halving with voltage is exactly what a fixed cable current does: the panel's wattage does not change, so halving the volts doubles the amps. 15 rows halve exactly; the 3 that do not are the odd 7-panel rows, where 3.5 is rounded up to 4 — the less conservative direction, and the one place the table argues with itself.
• It is a recommendation, not a maximum. Divide back out and every row lands between 9.7 A and 15.3 A at 220 V — all inside the 16 A a 2.5 mm² flex on a 16 A breaker gives you, most of them well inside. A computed maximum would sit at 16 A in every row. These are round numbers (7, 10 and 12 panels at 110 V, doubled) chosen to be safe, and the spread tells you they were quantised rather than calculated.

Where a brochure prints a panels-per-something figure, it is usually not this one. ROE Visual's spec tables give "Max. Hanging (panels)" and "Max. Stacking (panels)" — 12 and 5 for the Carbon series, 15 and 6 for Vanish, 13 and 6 for Black Quartz. Those are mechanical: how many tiles the top panel's frame can carry hanging beneath it, and how many can stand on the bottom one's without deforming it. They have nothing to do with the distro, and they get read as power limits constantly — partly because they sit two rows below the power-consumption line in the same table.

The mechanical limit is a real constraint and often the binding one for a column, so it is worth knowing both. It is simply a different question. Absen's own installation text is the honest summary of why the power figure is so rarely published:"the number of cabinets loaded on each power cable will be different upon different voltages and product models."

Put the two together for one common tile — a 500 × 500 at 200 × 200 px drawing 170 W — and the answer flips back and forth depending on nothing to do with the panel. At 8-bit on a 16 A European circuit the port binds; go to 12-bit and it binds harder; move to a 20 A American circuit at 120 V and the power binds instead. There is no general rule, which is the whole reason to compute both.

The panel's resolution and peak watts, the link and the signal format, the supply and the breaker. The tool gives the port capacity, the panels each line carries, and which of the two you will hit first.

• NovaStar, MX40 Pro LED Display Controller Specifications(V1.4.1) and the V1.5.1 revisions of the CX40 Pro, MX30 and KU20 sheets — the per-port capacities at 8, 10 and 12-bit that calibrate the whole data half, and the 9 Mpx device cap that binds before the ports do.
• Brompton Technology, Tessera Processor Output Port Capacitytable (dl.bromptontech.com, processor version 3.5.2) and the SX40 data sheet — the naive-packing counter-example, and a device cap that is a pixel rate rather than a bandwidth figure, so it does not move with bit depth.
• Absen, PL V2 Series User Manual, section 4.2 p.19 — the cabinets-per-power-cable table, and the accompanying statement that it depends on voltage and model. The same section states a per-port limit of 655,360 pixels, which sits just under the 659,722 the link arithmetic gives. Cabinet power fromPL V2 Series Specifications (usabsen.com, rev 2025-04-30).
• ROE Visual — Carbon, Vanish MkII and Black Quartz brochures (roevisual.com, 2024). Cited for what they donot contain: no power-chain figure and no connector type, but a Max. Hanging and Max. Stacking row that is read as one.
• NFPA 70 (NEC) 210.20(A) — the 80 % continuous-load rule used as the default derating. BS 7671 reaches a comparable place by way of cable rating and volt drop rather than a flat factor.

Assembled 21 September 2026 with AI assistance. The link arithmetic is checked at build time against NovaStar's three published MX40 Pro figures and reproduces all three exactly; the Absen table's two properties are asserted the same way. Those checks catch a transcription error and cannot catch a wrong source. Nothing here has been used to plan a wall that was then built. Sizing a circuit, and signing off a distro, is a job for someone qualified to do it.

Companion pages: cable voltage drop for the volt drop along the power chain this page ignores, Ethernet cabling for how far the data line will actually run on which category, and Line Calc, which does this arithmetic against a library of 150 real panels.