Deary me, I didn't think this whole thing would blow up into a huge argument. All I wanted to do was to let it be known there is a SUBTLE drop in sound pitch on the YMF289 and any ISA and PCI Yamaha audio chipset as those incorporate YMF289 cores. I'm not trying to IMPOSE which one is better, I'm just a picky bastard who will spot the smallest of differences in anything, especially audio, and I tend to gravitate towards the original design. It's the same thing with the Sega Genesis/MegaDrive, which originally used the YM2612 until Sega changed the FM chip to the YM3438 (the difference is much less subtle in certain circumstances as it's the overall output that sounds different between these two due to their altered DACs).

carlostex wrote:

He has none.

You want samples? You've got samples!

This also includes some OPL2 samples from the YM3812 on a 1990 revision of the AdLib. Also, the Vibra16 I used is VERY noisy (I was going to use a SoundBlaster Pro 2.0 CT1600, but decided not to because of its extremely heavy low-pass filter that the SoundBlaster 16s, YMF719 and AdLib don't have, which makes the sound much more muffled - it's my sound card of choice for DOS, but for consistency's sake, I left it out of this set of samples).

I will note that out of the games I sampled, Doom seems to make the lower sound pitch virtually impossible to hear. I can definitely hear it in X-Wing and a bit in TIE Fighter and Test Drive III, though it does seem some notes are more noticeably lower pitched than others. Depending on how sensitive your ears are, the difference may be so small, you can't even hear it.

Again, I just want to let the sound pitch difference be known. It's still genuine OPL3 (with a slightly lower sound pitch, but it's still genuine OPL3), and I will still recommend Yamaha YMF7xx cards as alternatives to a SoundBlaster, but for those who are INCREDIBLY anal like I am about even the most minute differences (minus low/high-pass filtering, that's personal preference - I prefer muffled FM Synthesis, which is why I like the SoundBlaster Pro 2.0's FM output over the SoundBlaster 16), I feel they should know about this so they can get a card with a YMF262 or CT1747. For those who are not as picky, by all means, get a card with a discrete YMF289 or an ASIC implementation of the YMF289 core.

Thanks for the new samples. Tried ABX'ing them with ReplayGain volume matching and wasn't able to tell a difference between CT2290 and YMF719 for myself, aside from some artifacts at the beginning of the recordings.

Also again, it's easier to find YMF289 and Yamaha designs with good sound quality than those with YMF262/CT1747. I'd like recommendations on nice SB16/AWE32 models.

I also wonder why words such as "inferior", "retarded", "trashing" are constantly being used to imply that another party is suggesting such things. Just amusing. 😀 I guess even words such as "for those who are not as picky" can provoke a reaction in some people. Hmm... Humans (myself included) can be so sensitive in many ways, don't ya think? Pun intended. 😀

archsan wrote:

distorted

I found a better word to describe it: "detuned". But only very slightly (so don't start making it sound like it's a bigger deal than it is please).

Yes, the YMF7xx cards are much easier to source. And they sound quite decent too. And the hanging note issues are enough to keep me from trying to find the needle in the SB16/AWE32 haystack. For users with ISA slots that aren't pursuing ultimate sound quality, I think it's worth it to put Aztech Sound Galaxy cards in your search list.

And CQM sounds readily discernible indeed, but that's outside the scope of this thread anyway ("OPL3").

[EDIT] I would still RECOMMEND Yamaha DS-XG cards (YMF724F-V's are what I have) heartily as a SB-Pro compatible, genuine OPL-3 PCI card (with XG as a bonus).

PhilsComputerLab wrote:

I just said that because X-Wing sounds awesome with Roland 😀

Of course! whenever Roland LA or SC or GM is available, that's a no brainer--even if I've only played the version with an early SB16 model some twenty years ago. I prefer something new to nostalgia. 😀 But then to you Roland must be the nostalgia I see, lucky you to experience its full glory back then. 🤣

Thank you for the recordings Ace!

firage wrote:

Thanks for the new samples. Tried ABX'ing them with ReplayGain volume matching and wasn't able to tell a difference between CT2290 and YMF719 for myself, aside from some artifacts at the beginning of the recordings.

Well done testing using volume matched ABX which is the only right way to compare audio being 100% bias free.
I can't hear any difference in pitch either, and I have a well trained musical ear.

People hear a difference (always better) in sound when changing power cords or speaker cables to more expensive ones, that's how strong subconscious biasing is. 🤣
Anything we "think" we ever so slightly hear is meaningless without an ABX test, it's a well known and used procedure in science when a "subjective prone" statistic is needed.

I suggest to anyone (especially people with perfect pitch) who claims he can hear the calculated/measured pitch difference of the YMF289 to do an ABX test.
Download Foobar200 with the ABX plugin.
Download Ace's comparison files.
Load one YMF262 and one YMF289 and do a 20-50 tries run.
Post a screenshot results here.
😀

carlostex wrote:

See?

And also hear. 😀 Check the pedia's article by the link with more attention. And you'll find the evidence, that 289 can play significantly different, what means clearly wrongly for some music made for 262.
Nuff said.

Ace wrote:

You want samples? You've got samples!

Thank you for providing these samples.

Again, I wasn't able to hear any differences 😒

I agree on the Sound Blaster Pro 2, it's my favourite "pure" OPL3 card as well.

Tertz wrote:

And also hear. 😀

Hear what?
In what sample, from where?

Tertz wrote:

Check the pedia's article by the link with more attention. And you'll find the evidence, that 289 can play significantly different, what means clearly wrongly for some music made for 262.
Nuff said.

The only thing this wiki page says about the sound is the following:

Wikipedia wrote:

However, the linear interpolator does not remove all artifacts, and the pitch of the output is subtly lower than the YMF262's as well.

Where did you get the assumption that the YMF289 is "significantly different" and "clearly wrong" is beyond me.
Nuff said?, quite the opposite, you have provided ZERO proof to back anything you stated.

The sample song in wikipedia is absolutely the worst you can compare by because it has down-bending notes, plus this statement "pitch of the output is subtly lower" right in your face.
The two combined can make anyone hear things like: unstable, detuned, wobbly, etc.. all created purely in ones mind.

PhilsComputerLab wrote:

Again, I wasn't able to hear any differences 😒

Why, did you expect to hear a clear difference of 0.39% (438.3 vs 440) in pitch between the two?
I don't think anyone will.

Hence why I said the difference is so subtle, there's a good chance you may not even notice the drop in sound pitch. In fact, when I listen back to the recordings, I notice the drop in sound pitch only with specific instruments.

archsan wrote:

I also wonder why words such as "inferior", "retarded", "trashing" are constantly being used to imply that another party is suggesting such things. Just amusing. 😀

That was never my intention, though sometimes, my wording may make it seem like I'm trashing something due to even the most minute difference. The fact is, the YMF289 is still a genuine OPL3 core with such a small drop in sound pitch, it may as well not even be there.

Also,

PhilsComputerLab wrote:

I just said that because X-Wing sounds awesome with Roland 😀

I hope you mean the SC-55 and not the MT-32; I think X-Wing sounds like shit with the MT-32. Honestly, I'd rather play this with FM, but that's mostly because I'm a sucker for FM Synthesis. 😜
Same deal with TIE Fighter (OPL3, not OPL2, thank you very much).

Tertz wrote:

carlostex wrote:

See?

And also hear. 😀 Check the pedia's article by the link with more attention.

I don't need. I use a mixer and i can compare my YMF-718 with a PAS 16 or any other YMF262 card in real time. It's very hard to distinguish which card is playing lower, due to the very small frequency difference. But I KNOW there is a difference, thank you. I've noticed this ages ago, and this is not a problem.

Tertz wrote:

that 289 can play significantly different, what means clearly wrongly for some music made for 262.
Nuff said.

Shut up.

James-F wrote:

Why, did you expect to hear a clear difference of 0.39% (438.3 vs 440) in pitch between the two?
I don't think anyone will.

Like i said before it is very hard to notice the difference when playing samples one after the other. However using a mixer, if you let a YMF card play ensemble with a YMF262 card, say a PAS 16, and you'll notice the 2 won't blend perfectly due to the frequency difference. That being said i don't give a shit the YMF cards play at 438Hz. OPL sounds great on YMF cards.

Tertz wrote:

that 289 can play significantly different, what means clearly wrongly for some music made for 262.

^I hope some people will understand why i used the word "retarded" before.

This thread concludes that Yamaha 71x cards with the YMF289 (OPL-L / OPL-SA3) have an excellent OPL3 chip on board but it operates at 438.3 to 439 HZ instead of 440Hz.
The difference is 0.39% percent or 0.067 semitones between the two and this difference cannot be heard by an average or musically trained person with normal use.

The Yamaha YMF71x cards have a very low self noise and are very compatible card for Dos or Windows gaming of the 90s.

James-F wrote:

This thread concludes that Yamaha 71x cards with the YMF289 (OPL-L / OPL-SA3) have an excellent OPL3 chip on board but it operates at 438.3 to 439 HZ instead of 440Hz.
The difference is 0.39% percent or 0.067 semitones between the two and this difference cannot be heard by an average or musically trained person with normal use.

The Yamaha YMF71x cards have a very low self noise and are very compatible card for Dos or Windows gaming of the 90s.

Agreed.

Whether or not the effect is large enough for the user to hear, this is still some fascinating science!

On the AWE32 front, it looks like the best liked three CT2760, CT3900 and CT3980 all work at the original sampling rate courtesy of their CT1747's. Interestingly, they don't have a 14.318 MHz clock source on the board - instead when it's present, the EMU chip is used to derive the clock using some algorithm.

firage wrote:

Interestingly, they don't have a 14.318 MHz clock source on the board - instead when it's present, the EMU chip is used to derive the clock using some algorithm.

Hmm, it makes me think if the designers of the YMF289 did a "close enough, nobody will hear it anyway" job to save some cash.

firage wrote:

Whether or not the effect is large enough for the user to hear, this is still some fascinating science!

On the AWE32 front, it looks like the best liked three CT2760, CT3900 and CT3980 all work at the original sampling rate courtesy of their CT1747's. Interestingly, they don't have a 14.318 MHz clock source on the board - instead when it's present, the EMU chip is used to derive the clock using some algorithm.

Or taken from the ISA clock which happens to be exactly that. 😀

stamasd wrote:

firage wrote:

Whether or not the effect is large enough for the user to hear, this is still some fascinating science!

On the AWE32 front, it looks like the best liked three CT2760, CT3900 and CT3980 all work at the original sampling rate courtesy of their CT1747's. Interestingly, they don't have a 14.318 MHz clock source on the board - instead when it's present, the EMU chip is used to derive the clock using some algorithm.

Or taken from the ISA clock which happens to be exactly that. 😀

I don't know what's wrong with using the ISA clock, but no, that pin is apparently not connected.

I have a feeling Yamaha may have used a 33.8688MHz crystal with the YMF289 due to 34MHz crystals being somewhat difficult to source. According to an earlier post by khokh2001, the clock divider divides the input clock by 684, and in this instance, if the 33.8688MHz crystal were to be replaced with a 34MHz crystal (not exactly the easiest thing to do as I've only found 4-pin oscillators that are 34MHz, not 2-pin like what's on the majority of sound cards), this would give a sample rate of just under 49708Hz, a measly 8Hz lower than the 49716Hz sample rate of the YMF262. At this sample rate, I don't expect any perceivable difference in sound pitch whatsoever (not that the YMF289's sound pitch difference is that big to begin with). However, I do wonder what this would do for the resampler. Might it fail to work properly with a higher input clock?

The crystal frequency on all later OPLs is selected so that the output sampling rate is exactly the standard 44.1 kHz (44100*768=33.8688 MHz) after resampling.
So if you change the clock frequency, you change the output sampling rate as well, so it could not be mixed with other sounds so easily without another resampling.
As also other stuff could be running at 44.1 as well, the sound chips could share the crystal, which is easy to find as the crystal frequency is very popular because of CD players.
Then again, for cost reasons, sound cards usually did not use 33.8688 MHz crystal but a less precise (0.3%) ceramic resonator.

Also each game might have their own values calculated for notes. Some datasheets used rounded frequencies like 3.58 MHz, 14.32 MHz, or even 3.6 MHz which would result exactly 50 kHz sampling rate. Therefore some MSX/OPL/OPL2/OPL3 datasheets say for A440 the note register should have 577 while 580 would be more accurate. And OPL4 datasheet says 582 because it had slightly lower rate for running the OPL core.

Working on getting some measurements in... I have a few problems with the test setup.

1. Many audio cards give me an output signal that's too low for some reason for my oscilloscope to process correctly... it sees the signal as being only 40-50mV for some reason, despite the sound being loud and clear when I commect headphones instead.

2. The average frequency it reads depends on the FFT size, with a threshold between 128k and 256k; below this it will read higher peaks, above it will read the peak as lower. Still trying to decide which is best to use.

...and another problem seems to be that the Adlib Tracker II which I was planning to use for the tests doesn't seem to actually work correctly with a real OPL3 chip! 😒

I does work with my Yamaha cards (YMF719, 724, 744) and MELT.AMD sounds great; it works correctly with a CS4237 card, an ESS1868, and even with the CQM on a SB32.

But when I try to use it with a CT2290 that has CT1747 on-board, or a Aztech Sound Galaxy II that has a real OPL3 chip, all I hear is garbage.
However if I use the OPL440 program I hear a clear 440Hz tone with both cards. Both cards also work great in games with AdLib music. But when I try to play AdLib music through the tracker with either of the 2 cards, all it puts out is random notes that have only a vague resemblance to the original tune. Moreover when I exit the tracker, the last sound played hangs and doesn't end. The same thing happens with both the CT2290 and with the Sound Galaxy.

stamasd wrote:

Working on getting some measurements in... I have a few problems with the test setup.
1. Many audio cards give me an output signal that's too low for some reason for my oscilloscope to process correctly... it sees the signal as being only 40-50mV for some reason, despite the sound being loud and clear when I commect headphones instead.

Consumer level audio (sound cards, MP3 players, etc..) will output 0.447V peak, so 40mV will be about -20db below peak.
This is very normal and expected.

http://www.sengpielaudio.com/calculator-gainloss.htm
https://en.wikipedia.org/wiki/Line_level#Nominal_levels