Three and a half volts

What the datasheet says

The MAX7219's Electrical Characteristics table has two rows about the signal pins, and they are absolute numbers rather than fractions of the supply:

• Logic High Input Voltage, VIH: 3.5 V minimum.
• Logic Low Input Voltage, VIL: 0.8 V maximum.

An Arduino Uno puts out 5 V and clears the first with a volt and a half to spare. An ESP32, an ESP32-S3, a Pico or a Raspberry Pi puts out 3.3 V, which is 0.2 V below it.

Between 0.8 V and 3.5 V the chip is not promising anything. In a real part the switching point is somewhere in the middle of that band — the datasheet lists a typical 1 V of hysteresis, which is what stops it chattering — so 3.3 V gets read as a HIGH by nearly every chip nearly all of the time. That is why the 3.3 V examples in the Lonely Binary repository work, and why the listing's "3.3 V or 5 V SPI signal compatible" is a fair description of what happens.

It is not what the specification says, and the difference shows up as intermittence rather than failure. Long jumper wires, a board running warm, a particular chip from a particular batch: any of those can move it, and when it moves you get a screen of garbage rather than a blank one, because the clock and the data are still arriving, just not as sent.

What to do about it

If you are on a 5 V board — an Uno, a Nano, a Mega — there is nothing to do.

If you are on a 3.3 V board and it works, it works. Keep the three signal wires short, keep them away from anything switching hard, and be aware that this is the first thing to suspect if the display starts misbehaving later.

If you want it in specification, put a 3.3 V to 5 V level shifter in DIN, CLK and CS. All three are push-pull signals driven in one direction only, which is the easy case: any of the boards in the logic level converter pack will do it, and the clock speeds here are nowhere near the limit of even the slowest of them.

Only the three signal wires need shifting. VCC still goes to the 5 V rail and GND is shared as before.

Why the chain is not affected

A chained board is not driven by your microcontroller — it is driven by the DOUT pin of the board before it, and DOUT is a 5 V chip's output. The datasheet puts it at V+ minus one volt when sourcing a milliamp, so a second board sees a 4 V HIGH against the same 3.5 V threshold.

That is worth knowing because it narrows the search. If board one is readable and board two is not, the level shifting is not the problem; the jumper, the device count or the power is.