[sdiy] Discrete MOS 6581 (Commodore C64 SID chip) schematics / analysis

[Rainer Buchty]

On Tue, 18 Aug 2026, Noah Vawter via Synth-diy wrote:

> there is not much remarkable about the SID except for its context, how 
> it is programmed with wavetables and the filter.

Which in this very context unfortunately also means modeling all the 
design quirks the various music routines relied upon.

1) 6581/8580 differences: combination-waveform output (leading to more 
useful combi-waveforms on the 8580), DC-offset removal (leading to quite 
silent §d418 samples on the 8580), and of course the general gain and 
cleaner sound (giving the 8580 more "punch")

2) phase accumulator and internal DAC behavior (or else the 8-bit sample 
technology won't work -- which it IIRC didn't with SID emulations 
predating the original "Fanta/SounDemoN" routine that brought us 
3-channel full-filterable and envelopeable 8-bit samples using a mere 
SID chip).

> It is 100% digital plus the pseudo analog filter. most of the 
> microcontroller implementers are lazy about the filter.

Which is a shame, because with ReSID there's a proven-working 
state-of-the-art C++ implementation (making VICE a faithful, authentic 
C64 emulator with all the bells and whistles, i.e., creatively used 
design quirks of the original).

You can tune ReSID to any 6581/8580 behavior to get the "right" sound 
for the used software.

> The ESQ-1 *is* the no-compromise SID synthesizer

I honestly never understood this comparison.

First of all, the SID is no synthesizer but a synthesizer voice :)

Second, while both, DOC and SID use phase accumulation at heart, the SID 
uses it directly, converting it into sawtooth (direct use), triangle 
(XOR the lower bits based on the topmost one), and pulse (like with 
analog designs, compare with pulsewidth value and either output max or 
min). DOC uses the phase accumulator solely for sample-memory access 
(and determining sync-out/osc stop).

> and predictable mass-produced filters (curtis? CEM?).

I wouldn't exactly call the CEM3379 "predictable", just check the 
datasheet for the horrible spread in voltage and resonance CV.

Hence, the ESQ1/SQ80 has software-based filter tuning (to at least get 
the frequency response somewhat right) and limits the resonance setting 
to a safe "will definitely not resonate" range.

> For example, people only rarely use LFOs in sid music.

Define "rarely". Most trackers do offer vibrato and PWM since the dawn 
of time (i.e., Chris Hülsbeck's SoundMonitor that's now turning 40), as 
it was paramount to let a single osc sound lively.

There usually also is a software-based ADSR (or just AR) envelope for 
filter modulation, sometimes also for PWM.

The program that put it to the max was the "Modulator" published in the 
German 64'er magazine back in (IIRC) 1985. It featured *eight* LFOs and 
a software envelope, where via the modulation matrix even the LFOs 
frequency and phase(!) could be modulated -- in addition to the obvious 
targets (osc frequency, pulse width, filter frequency). But it ate so 
many cycles that it could just about handle a keyboard scanner.

As for "recreations sound inferior", I think this is not true.

Here's Tuneful Eight (featuring an 8-SID stereo tune):

Real SIDs (8580):
 	https://www.youtube.com/watch?v=nhz3vHYX0E0

Too bad LMAN removed his very own FPGASID rendering, so let's use these:

UltiSID/FPGASID mix:
 	https://www.youtube.com/watch?v=KnFin5dxWLg

Pure UltiSID emulation:
 	https://youtu.be/SwN7anqqDP8

The FPGA-based ones IMO sound a tad harsher / cleaner / less "rounded", 
but only in direct A/B comparison. Probably not noticeable if heard 
alone. Adding a strategic tanh (approximation) here and there (or just 
some mild LPF at the output) might fix that already.

Considering the variances between the various 6581 revisions (and also 
C64 board revisions where the filter capacitors where changed from 
2200pF to 470pF -- or was it vice versa?) with the particularly audible 
change coming with the 6581(A)R4, not to mention the switch to 8580, 
makes discussions about "the" SID sound and emulating it "correctly" 
futile anyway, as there was never "the" SID.

If people today talk about "the" SID sound, there's at least two: the 
8580 for the "later" sound, and the 6581(A)R4 as the golden reference of 
the NMOS years.

Having said this, here's some nice (YMMV :) examples of current SID 
sound:

"Beat SID" (closing the circle, as the original was probably sequenced 
on a C64 using Steinberg's Pro16):
 	https://www.youtube.com/watch?v=fRYrSVIIz2o

"Robot - Not Human" (bringing its own video):
 	https://www.youtube.com/watch?v=CKi_m0NhwE0

"In Surgery" (way less minimalistic, yet nicely synced):
 	https://youtu.be/YbiGbyHKqy4?t=18

You want pure SID sample insanity? Ok, here's the latest work of Pex 
"Mahoney" Tufvesson, showing that there's always some cycles left to 
spare on other stuff (like the interleaved loader and decompressor plus 
a little bit of eye candy):
 	https://youtu.be/GqEI0T_WoAw?t=302

Citing: "This is still your unmodified Commodore 64 from 1982: 0.001GHz 
CPU, 0.000064GB RAM, 0.0002 billion transistors in 8000 nanometer 
technology, 0.0000002TB storage."

Cheers,
 	Rainer