Oof! Good question. I think most current OPL3 cores are pretty accurate.
But emulation of YAC512 DAC is another matter, maybe.
Also, Descent FM music on a SB16 CT1740 "noiseblaster" did sound different to an PAS16 performance, more bassy. That's the analogue side, though.
It's like with a Sega Genesis/MD Model 1 vs Model 2, I think.
The first version was very "full" or "heavy" sounding, but also noisy.

Any one that has Nuked, which includes 86Box.

I made UniPCemu, though that's OPL2. Right now the only cycle-accurate OPL2 that exists afaik.
Idk if cycle-accurate OPL3 exists, though.

Did add support for Sound Blaster Pro dual OPL2 recently. It sounds pretty great from what I've tested.

Idk how compatible dual OPL2 vs OPL3 is though.

superfury wrote on 2025-05-15, 16:52:

I made UniPCemu, though that's OPL2. Right now the only cycle-accurate OPL2 that exists afaik.
Idk if cycle-accurate OPL3 exists, though.

Do you have any recorded comparisons of your OPL2 emulation vs Nuked or other implementations?

GloriousCow wrote on 2025-05-15, 21:06:

superfury wrote on 2025-05-15, 16:52:

I made UniPCemu, though that's OPL2. Right now the only cycle-accurate OPL2 that exists afaik.
Idk if cycle-accurate OPL3 exists, though.

Do you have any recorded comparisons of your OPL2 emulation vs Nuked or other implementations?

Didn't do any. Simply went by ear comparing sound against youtube recordings (sounding exactly the same) and implemented based on reverse engineering reports of the chip on how it works (spoiler: it's a fully digital chip, not analog at all, all is performed using LUTs (quarter sinus wave, the various effects) and connecting logic (counters, single bit flips for negative sign and less documented logic connecting it all). The only part that's analog is done by the DAC connected to the chip, converting the 1-bit sign, 3 bit exponent and 10 bits mantissa floating point to a voltage on the output (in my case performed using a LUT for speed).
It started out as the simple one from fake86, but I pretty much rewrote the entire sound generating to a fully accurate OPL2 chip based on the reverse engineering instead.

All it basically has is a high-pass filter on the component that sends it's output to SDL, which is basically a 1Hz filter only, to prevent DC output.
Oh, and a 15392Hz low-pass filter after the DAC of each chip (left,right or mono(with pre-Sound Blaster Pro or just Adlib card emulation)). The setup (single/dual OPL2 or none) is detected by the emulated Sound Blaster(1.x, 2.o, Pro or none) combined with the Adlib emulation being enabled or not:
- No Adlib or Sound Blaster: no OPL2.
- Just Adlib: single OPL2.
- Pre-Pro Sound Blaster: single OPL2.
- Pro Sound Blaster: dual-OPL2.

The Game Blaster works much in the same way, except being split off onto a higher I/O base address (250h with full emulation of all non-SB chips as a standalone Game Blaster board) when Sound Blaster Pro is emulated, instead of the usual 220h base address on Pre-Pro being emulated.

My OPL2 has a bit more ringing in some cases though (feedback), perhaps because of the filters? Just with some songs (like MegaRace's) though. https://www.dropbox.com/scl/fi/0aodnll70emapl … t=ucwe8ia2&dl=0

The homework everyone copies:
AUDIO - Nuked OPL3 Patch for DOSBox(+Windows build) (SDL1)

There really hasn't been anything better than Nuked. There's just other loose end odd OPL3 clones to take on now (CQM etc)

leileilol wrote on 2025-05-16, 04:46:

The homework everyone copies:
AUDIO - Nuked OPL3 Patch for DOSBox(+Windows build) (SDL1)

There really hasn't been anything better than Nuked. There's just other loose end odd OPL3 clones to take on now (CQM etc)

Nukeykt's work on pretty much every one of his endeavors, at least the ones I have heard of and/or tried, has a habit of setting a high water mark for emulation accuracy.

superfury wrote on 2025-05-15, 16:52:

I made UniPCemu, though that's OPL2. Right now the only cycle-accurate OPL2 that exists afaik. Idk if cycle-accurate OPL3 exists […]
Show full quote

I made UniPCemu, though that's OPL2. Right now the only cycle-accurate OPL2 that exists afaik.
Idk if cycle-accurate OPL3 exists, though.

Did add support for Sound Blaster Pro dual OPL2 recently. It sounds pretty great from what I've tested.

Idk how compatible dual OPL2 vs OPL3 is though.

That's impressive ! I am more interested about OPL3 but it will interest a lot of people for sure !

leileilol wrote on 2025-05-16, 04:46:

The homework everyone copies:
AUDIO - Nuked OPL3 Patch for DOSBox(+Windows build) (SDL1)

There really hasn't been anything better than Nuked. There's just other loose end odd OPL3 clones to take on now (CQM etc)

Thank you, so there is no difference between emulators, it's just about this Nuked option which is here or not and which I saw only on dosbox-x

Yesterday I had a problem with dosbox-x who suddenly didn't want to take the F keys (which are obligatory for adlibtracker 2) it made MAC OS actions instead. It already happened and it's really annoying

Maz Hoot wrote on 2025-05-16, 08:38:

superfury wrote on 2025-05-15, 16:52:

I made UniPCemu, though that's OPL2. Right now the only cycle-accurate OPL2 that exists afaik. Idk if cycle-accurate OPL3 exists […]
Show full quote

I made UniPCemu, though that's OPL2. Right now the only cycle-accurate OPL2 that exists afaik.
Idk if cycle-accurate OPL3 exists, though.

Did add support for Sound Blaster Pro dual OPL2 recently. It sounds pretty great from what I've tested.

Idk how compatible dual OPL2 vs OPL3 is though.

That's impressive ! I am more interested about OPL3 but it will interest a lot of people for sure !

There was just one thing I couldn't figure out, though. There's two different hardwired modulators (each with two different speeds), one being the AM modulator and that other one. Managed to get the AM modulator working properly afaik, but had to adjust the other one based on Dosbox's code, as I couldn't find anywhere how it works in the OPL2 exactly. That's pretty much the only floating-point part in there right now.
Ah, it's the lfo_pm_table (vibrato) thing.
Basically it just manually recalculates the phase increment every sample, according to the PM wave LUT, which is indexed by the LFO phase.