PV string sizing, explained properly the full variable checklist behind NEC 690.7

Most of string sizing reduces to keeping the string inside a voltage window under the two most extreme temperature conditions the site will see. But there is a third constraint that sits alongside the voltage window and increasingly drives the design on modern high-current panels: the current each MPPT input can accept. Voltage sets how long a string can be. Current sets how many strings you can land on each MPPT. A complete check enforces all three.

The first two define the voltage window and are covered in the math below. The third defines how strings distribute across MPPT inputs, and it is the constraint that decides, for example, whether you run four shorter strings or three longer ones. We cover it in its own section because on today’s panels it is often the limit that actually binds.

The minimum is a hot-weather problem. Voltage falls as cells heat up, so the worst case for the lower bound is the hottest expected operating condition, when string Vmp is at its lowest and may drop below the inverter’s MPPT floor.

The maximum is a cold-weather problem. Voltage rises as cells cool, so the worst case for the upper bound is the coldest expected ambient temperature, when string Voc is at its highest and may exceed the inverter’s absolute maximum DC input voltage. NEC 690.7(A) specifically requires this cold-temperature Voc correction.

A valid string is any whole number of panels N where the minimum is satisfied at the hot extreme and the maximum is satisfied at the cold extreme. The job of a sizing check is to find that range precisely, using the right inputs, rather than approximating it and hoping.

The maximum DC input voltage is also a system voltage ceiling

The inverter’s absolute maximum DC input voltage is one ceiling, but it is not the only one. The maximum PV system voltage is also capped by the voltage rating of every component the string voltage touches: the conductors, the disconnects, and the inverter. NEC 690.7 sets the maximum permitted system voltage by application:

• One- and two-family dwellings: 600V DC maximum.
• Other installations such as commercial and multi-family above two units: commonly 1000V DC.
• Building exterior under specific 2023 NEC conditions (690.31): up to 1500V DC, with restrictions including 2000V-rated conductors and limits on entering buildings with habitable spaces.

The binding maximum for string sizing is the lower of the inverter’s absolute maximum DC input voltage and the system voltage ceiling for the application. On most residential string inverters the inverter ceiling binds first, but the application limit is a real constraint that a complete check respects, and exceeding it is a plan-check rejection even when the inverter itself could tolerate the voltage.

From the panel datasheet, a correct check reads eight values. Each one matters, and skipping or approximating any of them moves the result.

1. Vmp, voltage at maximum power point at STC. Drives the minimum-string calculation.
2. Voc, open-circuit voltage at STC. Drives the maximum-string calculation.
3. Isc, short-circuit current. Used downstream for wire sizing under NEC 690.8.
4. Imp, current at maximum power point.
5. Wattage, used for the DC/AC ratio and total system kW.
6. Vmp temperature coefficient, in percent per degree C, describing how Vmp falls with heat.
7. Voc temperature coefficient, in percent per degree C, describing how Voc rises with cold.
8. Pmax temperature coefficient, in percent per degree C, used only as a fallback when no Vmp coefficient is published.

The Vmp versus Voc coefficient trap

Here is a subtle error that a careless process makes and a rigorous one refuses to make. Vmp drops more steeply with heat than Voc does. If a tool substitutes the Voc temperature coefficient where the Vmp coefficient is missing, it understates hot-weather Vmp loss, which makes the minimum-string calculation optimistic and can yield a string that fails to start the inverter on a hot day.

A correct check does not silently substitute one coefficient for the other. If the datasheet only publishes a Pmax coefficient, the check uses that as the documented fallback and flags it back to the user rather than guessing. Transparency about which coefficient was used is part of a defensible design.

From the inverter specification, a correct check reads six distinct voltage constraints plus the per-MPPT current ratings. Many designers know three of the voltage values. The ones that get missed are the full-load MPPT voltages and the input current limits, and missing them is exactly what produces inverters that will not start on the hottest days or that fail from current overload on the brightest cold ones.

The six voltage constraints:

1. Startup voltage, the moment-of-truth threshold. String Vmp at hot temperature must exceed this for the inverter to wake at all.
2. MPPT minimum voltage, no-load.
3. MPPT minimum voltage at full load. This is often higher than the no-load minimum and is the binding floor on hot summer days, which is precisely when you can least afford to be near it.
4. MPPT maximum voltage, no-load.
5. MPPT maximum voltage at full load.
6. Absolute maximum DC input voltage, the hard ceiling that cold Voc times panel count must never exceed.

The two current constraints, per MPPT input:

7. Maximum operating input current per MPPT, the continuous current the MPPT will actually use. The combined Imp of all strings on that input must stay at or below this.
8. Maximum short-circuit input current per MPPT, the fault-level ceiling the input is rated to survive. The combined cold-corrected Isc of all strings on that input must not exceed it.

Note that constraints 7 and 8 are per MPPT input, not per inverter. A dual-MPPT inverter rated 13A per MPPT does not accept 26A on one input just because the other is empty. How many strings you parallel onto a single MPPT, and therefore how you distribute strings across the available inputs, is governed by these two numbers.

The binding minimum is the highest of three

For the minimum-string calculation, the binding constraint is the highest of the startup voltage, the no-load MPPT minimum, and the full-load MPPT minimum. Any string sized below that highest value risks an inverter that does not start. A tool that uses only the no-load MPPT minimum, ignoring startup and full-load minimums, will undersize the minimum string and ship a design that underperforms or fails to wake in real conditions.

This is the category that separates a correct sizing from a plausible-looking one. The string voltage window is defined by temperature extremes, and getting those extremes wrong invalidates the entire calculation.

The two temperature references

• ASHRAE 2 percent cooling design temperature, the hot-day reference for the minimum-string calculation. A US-temperate default is around 35 degrees C, but it is site-specific. Phoenix is closer to 45 degrees C.
• ASHRAE 2 percent minimum extreme low, the cold-snap reference for the maximum-string calculation required by NEC 690.7(A). A US default is around -2 degrees C, but again it is site-specific. Denver is closer to -16 degrees C, while Phoenix is around +1 degrees C.

NEC 690.7 points to ASHRAE data as the reference for these expected extremes. Using record-low temperatures instead of the ASHRAE design values is one common way designs end up either overly conservative or, worse, non-compliant when the wrong direction is chosen.

The mounting cell-temperature adder

This is the input that is missed most often, and it has a large effect. The panel cell runs hotter than the ambient air, and how much hotter depends on the mounting configuration and the airflow behind the module.

The implication is significant. A roof-mounted panel on a 35 degree C day is operating at roughly 65 degrees C cell temperature, not 35. That is where Vmp gets crushed, and it is why a string that looks fine if you forget the mounting adder can fall below the MPPT minimum in real summer conditions. The flush mount with the least airflow gets the hottest and is the most punishing for minimum-string sizing.

Here is the full calculation, in sequence.

Step 1: Hot cell temperature

T_cell_hot = T_ambient_max + mounting_adder

The hot cell temperature is the hot ambient design temperature plus the mounting cell-temperature adder from the table above.

Step 2: Hot Vmp per panel

Vmp_hot = Vmp_STC × (1 + tempCoeffVmp × (T_cell_hot − 25))

The Vmp at the hot cell temperature, derated from its STC value using the Vmp temperature coefficient. Because the coefficient is negative and the hot cell temperature is well above 25 degrees C, this value is meaningfully below the STC Vmp.

Step 3: Minimum panels per string

min_panels = ceil( mpptMinVoltage / Vmp_hot )

Where mpptMinVoltage is the binding minimum, the highest of startup, no-load MPPT minimum, and full-load MPPT minimum. The ceiling function is used because you need enough panels to clear the floor.

Step 4: Cold Voc per panel, required by NEC 690.7(A)

Voc_cold = Voc_STC × (1 + tempCoeffVoc × (T_ambient_min − 25))

The Voc at the cold ambient extreme, increased from its STC value using the Voc temperature coefficient. Because the coefficient is negative and the cold ambient is below 25 degrees C, the term increases Voc above its STC value. This is the cold-temperature correction NEC 690.7(A) specifically requires.

Step 5: Maximum panels per string

max_panels = floor( maxDcVoltage / Voc_cold )

Where maxDcVoltage is the inverter’s absolute maximum DC input voltage. The floor function is used because exceeding the ceiling even once damages the input, so you round down.

Step 6: The valid range

The valid string size is any integer N such that:

min_panels ≤ N ≤ max_panels

If min_panels exceeds max_panels, no valid string exists for that panel and inverter pairing at that site, and the equipment selection itself has to change. A good check surfaces that immediately rather than letting the designer discover it later.

A fully worked example

Take a real residential pairing on a standard roof mount in Denver.

• Panel: 410W module. Vmp 31.2V, Voc 37.4V at STC. Vmp temperature coefficient -0.35 percent per degree C. Voc temperature coefficient -0.25 percent per degree C.
• Inverter: 600V absolute maximum DC input. Binding MPPT minimum (the highest of startup, no-load minimum, and full-load minimum) of 80V.
• Site, standard roof mount under 5 inches: ASHRAE cooling design 33 degrees C, ASHRAE extreme low -16 degrees C, mounting adder +30 degrees C.

Minimum string, hot side:

T_cell_hot = 33 + 30 = 63 °C
Vmp_hot = 31.2 × (1 + (−0.0035 × (63 − 25))) = 31.2 × (1 − 0.133) = 27.05 V
min_panels = ceil( 80 / 27.05 ) = ceil(2.96) = 3

Maximum string, cold side, per NEC 690.7(A):

Voc_cold = 37.4 × (1 + (−0.0025 × (−16 − 25))) = 37.4 × (1 + 0.1025) = 41.23 V
max_panels = floor( 600 / 41.23 ) = floor(14.55) = 14

Valid range: 3 to 14 panels per string. A designer would typically land near the top of that range, perhaps 12 or 13, to keep the operating voltage high in the efficient part of the MPPT curve while leaving margin below the 14-panel ceiling.

Now watch what a single dropped input does. Forget the +30 degree C mounting adder and compute the minimum at 33 degrees C instead of 63 degrees C, and Vmp_hot reads 30.3V instead of 27.05V, which still gives a minimum of 3 here but in a colder climate or against a higher MPPT floor would understate the minimum and let through a string that will not start on a hot afternoon. Substitute a shallower Voc coefficient of -0.20 percent for the real -0.25, and Voc_cold reads 40.4V instead of 41.23V, which can push the apparent maximum from 14 to 15 panels and produce a cold-morning overvoltage that exceeds the 600V ceiling. The math is simple. The discipline of using the right inputs every time is the hard part.

The voltage math tells you how long a string can be. It does not tell you how many strings you can run or how to distribute them across the inverter’s MPPT inputs. That is a separate decision governed by the per-MPPT current limits, and on modern high-current panels it is frequently the constraint that actually binds the design.

Why this is now the binding constraint

Panels have gotten dramatically more powerful, and the way they got there is partly higher current. A current-generation module can have an Imp above 13A and an Isc higher still. Meanwhile, a typical residential string inverter MPPT input is rated around 11 to 16A operating. The arithmetic gets tight fast. One string of high-current modules can sit comfortably on an MPPT. Parallel a second string onto the same input and the currents add, and you can blow past the MPPT’s input rating even though the voltage window is perfectly satisfied.

This is the failure mode that voltage-only sizing misses entirely. A string can pass every voltage check and still overdrive the MPPT on current. The inverter may run through a mild season and then fail on the first bright, cold day, when current is at its highest. The cold-temperature correction that raises Voc also raises Isc, so the worst case for current is the same cold, sunny condition that sets the maximum voltage.

The design tradeoff: fewer longer strings versus more shorter strings

Here is where it becomes a real design lever rather than a pass-fail check. Suppose you have 48 panels to wire onto a dual-MPPT inverter, and the per-MPPT current rating only comfortably accepts one string of these high-current modules per input.

• Four strings of 12 would require paralleling two strings onto each of the two MPPT inputs. That doubles the current on each input and can exceed the MPPT rating.
• Two strings of 24, if 24 panels fit inside the voltage window, puts one string on each MPPT, keeping each input within its current limit.

The total panel count is identical. The total array power is identical. The difference is entirely in how the strings are configured and distributed, and only one of the two configurations is valid on the current constraint. The voltage math alone would happily approve both. This is exactly the situation where four shorter strings will not work but three longer ones, or two longer ones, will.

The general principle: when the per-MPPT current limit is tight, you favor fewer, longer strings to reduce the number of parallel currents landing on each input, as long as the longer strings still fit under the cold-Voc maximum. The voltage window sets the ceiling on how long you can go. The current limit sets the floor on how few strings you can get away with. The valid design lives where both are satisfied at once.

What a complete check does

A complete check does not just validate one string in isolation. It looks at the whole array and the inverter’s MPPT topology together, and it verifies:

1. Each string length is inside the voltage window, from the math above.
2. The number of strings assigned to each MPPT input, multiplied by the per-string cold-corrected Isc, stays within that input’s short-circuit current rating.
3. The combined Imp on each input stays within the operating current rating.
4. Where the current limit forces a choice, it steers toward the string configuration that distributes current legally across the available inputs, which often means fewer and longer strings rather than more and shorter ones.

This is why string sizing and string distribution cannot be fully separated. The length of the string and the number of strings per MPPT are solved together, against the voltage window and the current limits simultaneously. A tool that sizes the string for voltage but never checks the resulting current per MPPT will approve configurations that smoke the inverter.

For microinverter systems, the entire voltage-window framework above does not apply, because the voltage conversion happens at each module. The binding constraint is not voltage. It is branch circuit current.

A correct microinverter check looks up different values:

• Maximum units per branch, keyed by breaker size. The limit differs for a 15A, 20A, or 25A branch breaker, and it comes from the device’s branch circuit specification map.
• Default breaker size for the device.
• Branch phases, single-phase versus three-phase Wye, which determines whether the trunk cable is two-conductor or three-conductor.
• Branch line-to-neutral specification, which determines whether a three-phase trunk needs an unbalanced neutral.

The maximum number of microinverters per branch is then:

max_units = min( max_units_per_branch[selected_breaker], roof_count_limit )

This is governed by NEC 210.20 for the branch circuit overcurrent protection and NEC 690.8(D) for the system. A tool that applies string-voltage logic to a microinverter system, or branch-current logic to a string system, is applying the wrong constraint entirely.

A sizing check that only returns pass or fail on the minimum and maximum is incomplete. Several conditions are technically valid but signal a suboptimal or risky design, and a rigorous check surfaces them.

High DC/AC ratio warning. When string DC watts divided by inverter AC nominal exceeds 1.5, clipping becomes likely, and the check should issue a clipping advisory. A modest DC/AC ratio above 1.0 is normal and often beneficial. An aggressive one wastes potential production.

Operating voltage deviation warning. When string Vmp at STC is more than plus or minus 20 percent off the inverter’s nominal DC operating voltage, the string is operating at a suboptimal point on the MPPT curve, and the check should flag it. The string may be valid but is not where the inverter is most efficient.

Missing temperature coefficient warning. When the panel record has no Voc temperature coefficient, a rigorous check refuses to substitute a generic silicon default such as -0.27 percent per degree C. It fires a panel-Voc-coefficient-unknown warning instead. Silently using a default here can produce an NEC 690.7(A) violation if the real coefficient is steeper than the default.

Missing project temperature warning. When the project’s environmental configuration is empty, a rigorous check refuses to size using default temperatures. Sizing a cold-climate project with mild default temps would ship NEC 690.7(A) violations, because the real cold extreme produces a higher Voc than the default assumed. The check should stop and require the site temperatures.

The theme across all four warnings is the same. The dangerous failures in string sizing are not the loud ones that obviously fail. They are the quiet ones where a default gets substituted, the number looks reasonable, and the violation only appears in the field or in a plan-check rejection.

The resolved string size is not the end of the process. It feeds several downstream calculations, and an error in the string size propagates into all of them.

DC string wire sizing. The minimum conductor ampacity is Isc times 1.25 for continuous use per NEC 690.8(A)(1), times another 1.25 for the overcurrent device safety factor per 690.8(B). Those two 1.25 factors compound, and the conductor must carry the result.

Small-conductor overcurrent bump. NEC 240.4(D) caps the overcurrent protection relative to the conductor AWG to prevent fault-current problems in small conductors. This interacts with the wire sizing above.

PV wire stocking floor. Industry convention sets a number 10 AWG minimum for DC string wires, because number 12 PV wire spools are rare in practice. A correct process respects the practical stocking floor, not just the theoretical minimum gauge.

String-to-MPPT assignment. When multiple strings share one MPPT input, they must have matching panel counts, or close enough that the resulting voltage ranges overlap for both strings. Mismatched strings on a shared MPPT operate at least one of them off its true maximum power point.

Maximum input current per MPPT. Covered in its own section above, because it is a design constraint rather than just a downstream calculation. The resolved string count and distribution must keep the combined cold-corrected current on each MPPT input within that input’s operating and short-circuit ratings. On modern high-current panels this is often the limit that decides the final string configuration.

Part of getting string sizing right is knowing what does not affect it, because conflating these with sizing inputs wastes time and sometimes produces wrong designs.

Roof azimuth and tilt. These affect energy production, not string sizing. A south array and a west array of the same panels on the same inverter size identically. Orientation changes how much energy the string makes, not the MPPT voltage window it has to live within.

Shading. Shading affects per-string output and is a real production and layout concern, but it does not change the MPPT envelope the string must satisfy. The voltage window is set by temperature and equipment, not by partial shade.

Wire run length. Run length affects voltage drop and therefore wire gauge selection, not the number of panels in the string. A longer run may need a heavier conductor, but it does not change the valid string size.

Inverter brand preference. Every inverter’s MPPT window is a fixed electrical property. There is no brand preference that changes the math. The window is what it is, and the string either fits inside it or does not.

Treating any of these four as a string-sizing input is a sign of folklore rather than method.

Read back through the preceding sections and count the inputs: eight panel values, six inverter constraints, three environmental references including a mounting adder that depends on standoff height, two compounding temperature corrections, a branch-current path for microinverters, four warning conditions, and four downstream propagations. Every one of them has to be correct, sourced from the right datasheet field, and applied in the right direction.

This is exactly the kind of work that is error-prone by hand and reliable in software. Not because designers cannot do the math, but because doing it correctly for every string on every project, with current datasheet values, the right ASHRAE temperatures for each site, the correct mounting adder, and a refusal to substitute missing coefficients silently, is a discipline that a spreadsheet does not enforce and a tired designer at the end of a long day does not always maintain.

The failure modes are quiet. A substituted coefficient, a forgotten mounting adder, a default temperature on a cold-climate site. Each looks fine on the page and shows up later as an inverter that will not start, a plan-check rejection citing NEC 690.7(A), or a cold-morning overvoltage event. The cost is not the math. The cost is the rejection cycle, the truck roll, and the warranty argument.

This is the part of design that Solar Design Lab automates. Our designer runs the full check above on every string you place: it reads the eight panel values and six inverter constraints from the equipment records, applies the site-specific ASHRAE temperatures and the correct mounting adder, runs both temperature corrections, refuses to substitute missing coefficients silently and warns you instead, fires the DC/AC and operating-voltage advisories, and feeds the resolved string size into the downstream wire and OCPD sizing. The point is to make the correct, code-compliant result the default outcome of placing a string, so the quiet failures never reach the field or the plan reviewer.

For designers evaluating tools, this is the level of rigor worth checking for. Ask whether a tool exposes the full-load MPPT minimums, whether it applies a mounting cell-temperature adder, and whether it substitutes missing temperature coefficients silently or warns you. Those three questions separate tools that look like they size strings from tools that actually do.