Chiptune Nostalgia and the Real Technical Constraint

Chiptune has been a going concern for about twenty-five years now — 8bitpeoples set up shop in 1999, Blip Festival ran in New York from 2006, LSDJ turned a Game Boy into a gigging instrument, and there is a steady supply of new music that sounds like a machine I owned. I like a lot of it. I also think the scene has inherited the wrong half of the thing it admires, and the mistake is specific enough to be worth naming.

The common account goes: old sound chips had a small palette of harsh waveforms, and composers made beautiful things inside that palette, and now we reproduce the palette as homage. Square waves, sawtooth leads, noise-channel snares. Timbre as heritage.

The palette was the least interesting restriction on the machine.

What was actually squeezing

Three things pressed on a game composer in 1986, and none of them was the waveform list.

Channel count against arrangement. The SID has three voices. The Game Boy has four. The NES’s 2A03 has five, one of which is a sample channel with a tight memory budget. A conventional arrangement wants bass, harmony, melody and percussion, so on a C64 you are already one short before anyone has hit anything. The famous C64 arpeggio — a chord played as its constituent notes cycling at fifty times a second until the ear fuses them — exists because three is less than four. It is arithmetic wearing a costume. I’ve gone through the mechanics of that in The C64 SID Chip and Music as System.

CPU time against the game. The music was a subroutine. It ran on a raster interrupt and it had a budget measured in scanlines, and every cycle it spent was a cycle the game logic didn’t get. A composer who wanted continuous pulse-width modulation under a lead was asking the programmer to give up sprite multiplexing somewhere. That is a negotiation between two people about a shared resource, and it happened on every project.

The sound effects, which had priority. Here is the one nobody reproduces. The game needs to make a noise when you shoot, and there is no spare channel, so the music gives one up. A voice drops out mid-phrase, the laser plays, the voice comes back. Composers wrote around this deliberately — putting the least structurally important line on the channel most likely to be stolen, writing bass parts that survive having their tail cut off, designing phrases with holes in them because holes were coming whether they were designed or not.

That last one is the deepest constraint of the three, and it is the one that says something true about the medium: the music was a component in a real-time system, sharing hardware with the thing the player was actually doing. It was never a track. It was a process, competing.

The tells, and where they came from

Take the sonic signatures of the era one by one and they all trace back to those pressures rather than to any waveform.

The arpeggio: channel scarcity. The noise-channel snare with a bass note briefly vanishing underneath it: channel scarcity again, resolved by scheduling. The pulse-width modulation that makes a single voice sound like two slightly detuned ones: a CPU-cheap way of buying thickness you couldn’t afford in voices. The tiny click before a C64 bass note: a hard restart, working around a genuine bug in the 6581’s envelope generator. The way Rob Hubbard’s basslines have that particular restless motion — and there’s a longer argument about his authorship — is partly a man keeping a single channel interesting because he has no second one to bring in.

Every one of those is a solution. Modern chiptune reproduces them as vocabulary. You can hear an arpeggio in a track running eight channels in a tracker with no game attached and no interrupt budget and nothing to duck for. It sounds correct. It means nothing, in the sense that it isn’t answering anything.

Consider the loading-screen tunes, which are the strangest artefact the era produced. A tape took several minutes to load, so composers wrote music to fill dead time on a machine doing almost nothing else — which meant, for once, nearly the whole CPU. Those tunes are consistently more elaborate than the in-game music sitting on the same cassette, and the reason is budgetary rather than artistic ambition: the loader wasn’t running a game, so the interrupt was free. Players sat through loads to hear them finish. That is a direct measurement of what the frame budget was costing the rest of the soundtrack, and it’s visible only because the era accidentally ran the control.

That isn’t a moral failing, and I want to be careful here, because “your music doesn’t suffer enough” is a stupid thing to say to a musician. Style is legitimate. Surf guitar doesn’t require a Fender Reverb tank from 1963 to be good surf guitar. The point is narrower: when we talk about what 8-bit music achieved, we tend to praise the timbre, which was free, and skip the scheduling, which was the work.

Real constraint is still on the table

The interesting corner of the scene is the part that takes the hardware seriously enough to lose arguments with it.

LSDJ is a Game Boy cartridge. Johan Kotlinski shipped it in 2000, it runs on the actual DMG, and it gives you four channels because the DMG has four channels. Nanoloop is the same bargain. People gig with these, and the ceiling is real — when a Nullsleep set sounds dense, the density was constructed inside a limit that would have stopped it if the construction failed. Same with the FamiTracker end of things, where the rules are the 2A03’s rules and the expansion chips are the ones Konami and Sunsoft actually shipped in Famicom cartridges.

Jake Kaufman’s Shovel Knight soundtrack is the clarifying case, because he documented the compromise. He wrote it in FamiTracker to Famicom rules and permitted himself the VRC6 expansion — a real chip, in real 1988 cartridges — then broke his own rules in a handful of specific, listed places. That’s a musician being explicit about which constraints are load-bearing and which are costume, which is more intellectual honesty than most of the genre manages.

And then there’s Anamanaguchi playing a hacked NES through a band, where the constraint and the escape hatch are on stage simultaneously and you can hear the seam. That’s the whole argument in a live rig.

What these have in common is that the hardware can say no. A DMG will refuse a fifth voice. FamiTracker under 2A03 rules will refuse a chord you haven’t arpeggiated. The refusal is the creative partner, and everything the scene calls characterful turns out to be the mark left by an argument someone lost and then worked around. Take the refusal away and you have a preset.

The Amiga is the control experiment

If you want proof that the squeeze was the point, look at what happened when it lifted.

Paula gave the Amiga four channels of 8-bit sampled audio, DMA-driven, so the CPU cost of playback collapsed and the palette went from “four waveforms” to “any sound you can record”. Karsten Obarski’s Soundtracker arrived in 1987 and the module scene followed. Suddenly a composer could use a real snare drum. And the music changed character immediately — the arpeggios thinned out, the scheduling acrobatics went away, and the craft moved to sample selection and memory budgeting, which is a different craft with different virtues.

Amiga module music is frequently excellent. It also sounds like a different discipline, because it is one. Nobody had to steal a channel from the bassline for a laser any more. The demoscene took the freed capacity and ran somewhere else entirely with it — a lineage worth its own piece — and the specific thing that made SID music feel like a mechanism quietly stopped being available.

What the nostalgia is for

I think what people are actually responding to, when they respond to this stuff, is audible effort. You can hear someone solving a problem in real time. A C64 tune has the shape of its own difficulty in it — the ducking, the flicker, the fused chord, the click — and that legibility is rare in music. Most production hides the work. This production couldn’t.

Which suggests the useful move for anyone making chiptune now: pick a constraint that costs you something. Four channels and mean it. A CPU budget. Ducking for effects that don’t exist, imposed as a rule. Something that can beat you. The waveforms were always free, and free things don’t leave marks.

The best of the old catalogue holds up as music with the hardware context stripped away entirely — put a Hubbard tune in front of someone who has never seen a breadbin and it still moves. But it holds up because of what the hardware did to it, and reproducing the surface without the pressure gets you a fair likeness of the effect with nothing underneath. Three voices was interesting because three voices was not enough. That was the whole gift.