Pull a nozzle check from an i1600 machine and one from an i3200 machine and lay both sheets on the same table. Same print width, same ink, same droplet sizes on the spec sheet, and the second sheet has twice as many blocks of lines. That is the entire difference between these two heads, and it is worth understanding before you pay for either one.

They are the same block. 69.1 by 59.5 by 35.6 mm, 80 grams, 33.8 mm of effective print width, four colors, 43.2 kHz. Epson did not build the i3200 bigger. Epson put two nozzle rows per color where the i1600 has one, and offset them by half a pitch.

So the question is never which head is better. It is whether your machine can turn the extra rows into meters, and whether your shop runs out of something else first.

You can answer both with a stopwatch. That is what follows.

Same Block, Twice the Nozzle Rows

Epson lists both heads on one page of the same document, and that is the first useful fact: width, depth, height, weight, print width, color count, firing frequency and droplet sets match. The i1600 carries 1,600 nozzles in four rows, one row per color, 400 nozzles per color. The i3200(4) carries 3,200 nozzles in eight rows, two rows per color, 800 nozzles per color.

The two rows of a color are not stacked one behind the other. They are offset by half a nozzle pitch and interleaved, which is why the i3200 lays down 600 npi of a single color in one pass while the i1600 lays down 300 npi in one pass. Every practical difference between these two heads comes out of that one change in geometry.

Note what did not change. A single nozzle on an i3200 fires no faster than a single nozzle on an i1600. Epson gives 43.2 kHz for single dot mode on both, and the same three level droplet sets, 3.8, 6.2 and 9.3 picoliters on the UV versions. The head is not quicker. There are twice as many of them working on the same inch of film or vinyl.

Count the blocks before you argue. Print a nozzle check and count blocks of lines, not individual lines. Four blocks is a 1,600 nozzle head, one per color. Eight blocks in four pairs is a 3,200 nozzle head. Thirty seconds, no tools, and it settles most arguments about what a used machine has inside, including machines sold as upgraded where the original head is still fitted.

One ordering trap. Epson also sells a part labeled i3200(8). It holds 3,200 nozzles as well, but they are spread across eight colors at 1,200 npi on four rows, and the body is 1.1 mm taller than the four color version. It is a different part for a different ink set. Order by the string i3200 alone and you can receive something that will not run four color CMYK at all. Order with the full suffix, A1, E1 or U1, and with the row count confirmed in writing.

Keep the two questions apart when you order. The suffix tells you the ink chemistry, A1 for aqueous, E1 for eco solvent, U1 for UV, and both heads exist in all three. The row count tells you the density. That is six parts behind two familiar names, and the suffix matters more than the number: a head fitted for the wrong chemistry usually seals and fails within weeks, while a head with the wrong row count tends to refuse to work at all. Epson's own viscosity figures mark the split, 5.0 to 7.0 mPa·s on the UV version against 2.2 to 4.5 on the aqueous and eco solvent versions.

There is a second difference that shows up after a year of daily work. Eight rows per head means twice the nozzle lines to scan every morning, and twice the surface area where a partial clog can start. On a machine where somebody actually reads the sheet, that is ten extra seconds. On a machine where nobody reads it, it is a doubled chance of running all afternoon on a head that was already half gone.

Why Twice the Nozzles Is Not Twice the Output

The pass arithmetic is easy on a napkin. A job that needs 600 npi of coverage per color takes one pass from a head that interleaves two rows and two passes from a head that does not. The i3200 finishes that file in roughly half the carriage passes, and on a machine built to feed it, that is where the speed comes from.

Built to feed it is the part that gets skipped. Eight rows at 43.2 kHz is twice the pixel rate of four rows. The mainboard, the head board and the flat cables have to carry that without dropping data. A controller designed around four rows will refuse the head, print only half the rows, or quietly lower the firing frequency until your new head runs at the speed of the old one. This is the real reason the same i3200 part number turns in very different numbers in a cheap chassis and in a production machine.

Downstream of the head, the ink has to go somewhere. Twice the dots per second is roughly twice the ink per second. On eco solvent that lands on the pre, print and dry heaters and on the airflow in the curing zone. On UV it lands on lamp dose per unit of travel, and white ink is the first to show it, because titanium dioxide already blocks the light. On DTF it lands on the powder station, where a wetter, heavier film takes up powder unevenly and cures with a crust that cracks at the edge of the transfer.

Mechanically there is one piece of good news worth knowing. Both heads weigh 80 grams, so fitting the denser head adds no mass to the carriage and needs no stiffer belt. What it does do is run the carriage faster for the same file, and higher traverse speed magnifies the errors you already have. A slightly worn belt, a dirty encoder strip or a lazy damper shows up as visible banding much sooner at production speed than it did at the speed you were running last month.

Time the whole chain, not the printer. Print one meter at the pass count you actually sell, then follow that meter through powder, cure and press with a stopwatch. Write the four numbers on a card and tape it to the machine. If printing is the largest number, a head with more rows is worth pricing. If it is not, you have just saved the price of a head with ten minutes of work.

There is a cheaper way to find your ceiling. Take a file you print every week and run it one pass count lower than your normal setting. If the machine holds quality there, your current setting has headroom and the head is not what limits you. If it falls apart, the head is already doing real work and more rows would buy something. Either answer costs one meter of film and twenty minutes.

The judgment sentence. If printing is already less than half of your per meter time, halving print time buys you less than a quarter of the total, and the head is the wrong purchase this month.

What a Head Swap Can and Cannot Do

Same footprint does not mean drop in. The mounting line and the print width match, which fools people into thinking the swap is mechanical. It is electrical and it is firmware. Eight rows against four means twice the data channels and a different drive requirement, and the machine has to support the specific head model on all three levels, mechanically, electrically and in firmware.

Some Chinese builders ship the single head i1600 kit and the single head i3200 kit on the same mainboard and the same head board, changing only the firmware build. That claim comes from the kit sellers, so treat it as a question to put to your supplier, not as a rule. It does explain a real pattern: two machines that look identical inside can behave completely differently because one controller carries an i3200 build and the other does not.

What a swap cannot do is add print width. The effective width of both heads is 33.8 mm. If the hope was that a head change would let you cover a wider roll in fewer passes, it will not. Head width sets the pass overlap on a scanning machine, and pass overlap sets how many times the carriage crosses your media.

Run these five checks before you order.

  1. Read the head model string in the service menu or in the RIP, not the label on the box the last head arrived in. If the string and the physical head disagree, somebody has been in there before you and you are inheriting their decision.
  2. Photograph the old head label with the full suffix readable. A1, E1 and U1 are one character apart, and two of those three will damage themselves on the wrong ink.
  3. Ask the machine builder, in writing, which head models the current firmware build supports. A vague answer here is the cheapest warning you will ever get.
  4. Confirm the head rank and ID write procedure for your model and who will perform it. A replacement head that prints without its rank written looks fine for a week and then starts shedding nozzles.
  5. Check manifold and damper fit. The i3200-A1 ships with a manifold, and your existing damper stack has to match it, or you will be chasing air in the line for a month.

If the machine accepts the head but prints only half the rows, or throws a head detect error at power up, stop. That pattern is a data path problem, not a head problem, and fitting a second new head into the same socket is the most expensive way to learn it. Put the old head back, confirm it still prints, and take the question to the builder with the model string and the firmware build number in hand.

Price the decision properly. A head swap costs the part, the freight and the hours the machine stands still. Downtime on a working line is the output you did not make, plus the film and powder scrapped mid roll, plus the expedited freight, plus a second visit if the rank write was missed. A swap that ends with the machine printing at exactly the old speed has spent all of that for nothing, and that is the usual outcome when the head was never the limit.

Which Machine Should Get Which Head

Two one line judgments cover most of it. If you run sampling, short runs and a mixed order book, the i1600 is the safer head, because the part costs less, the maintenance curve is gentler and a mistake costs less output. If you run long production shifts on one product with an operator who checks nozzles every morning, the i3200 pays for itself in passes.

Between those two there is a third case, and it is the one that gets missed: the shop where the printer is not the constraint at all. On DTF that is a shaker and a press that cannot keep up with the printer. Adding nozzle rows to the printer moves the queue, it does not remove it.

Maintenance asymmetry runs the opposite way from what most buyers expect. The i3200 has 800 nozzles per color against 400, so there are more nozzle lines to read and more places for a partial clog to begin. It also covers a color band in fewer passes, which means one dead nozzle has fewer chances of being filled in by a later pass landing on the same line. A dropout that a two pass i1600 job would average away shows up as a visible line on a single pass i3200 job. That is not a defect in the head. It is what parallelism looks like on paper, and it makes the morning check non negotiable on that machine.

Stock the same suffix you run. A U1 sitting on the shelf as a spare for an A1 machine is not a spare, it is a future destroyed head and a delayed order. Write the suffix on the shelf label, not on an invoice in a drawer.

There is a third configuration worth understanding before you compare quotes: two i1600 heads staggered on one carriage, both carrying the same four colors. The rows of the two heads interleave the way the i3200 does internally, and a machine set up that way reaches comparable coverage from two cheaper parts. What you gain on the parts you pay back in alignment, in two dampers per color, in twice the nozzle checking, and in the stagger between the heads, which drifts and has to be redone. One i3200 is a denser part. Two i1600s are two parts that have to agree with each other every morning.

One more thing worth doing before any of this. Walk the shop at the end of a busy day and note what people are waiting on. If they are waiting on the printer, the nozzle count question is real. If they are waiting on the press, on the powder station or on the trimming table, then the extra rows are a number on a spec sheet that will never reach your invoice in a good way.

And if you cannot finish this sentence with a number in it, you are not ready to buy: the thing that limits my output is ______, and it costs me ______ meters a shift. Fill both blanks and the choice between 1,600 and 3,200 nozzles makes itself.