I wish some games that did use ADPCM could do it through hardware.

Quake3 actually had a few versions with ADPCM sound output but it ended up being slower (s_compression 2) and it was removed in later versions. Think it was in 1.27g at the latest

Only other ADPCMey Windows games I can think of are Thief, System Shock 2, GTA2 and Max Payne and I don't believe those did it through hardware either. It's more of memory savings and/or disc streaming than anything else (and the loss works out in GTA2's favor for more radioqualityness).

486 DX4 100 MHz is capable of decoding .mp3 at 22050 kHz stereo *or* 44100 mono. Basically, at half the output data. The CPU load is around 100%.
.mp2 files require much lower computational power, but higher bitrate for the same quality (I'd say a 160 kbps mp2 is equivalent to 128 kbps mp3).

tpowell.ca wrote:

Since all soundcards of the day with PCM output supported 8bit unsigned mono 8kHz and 11kHz samples, that was what most used with the exception of the odd ducks that used soundcard-specific features (eg: Duke Nukem 2).

Well, in most cases with early games, the SoundBlaster was the only device that was supported for samples anyway. So it was already a soundcard-specific feature. Since all SoundBlasters also support ADPCM, it doesn't really matter whether you use uncompressed or compressed samples.

True, but ADPCM could afford you the ability to have perceptually better then 8bit resolution at only 4bit-equivalent sample sizes.
So, developers could have recorded sounds in 22kHz ADPCM (from 16bit source material for highest benefit) and it would take the same space as 11kHz 8bit. That would have been a great feature.

Problem is that it is not possible to mix two ADPCM sources without decompressing them first, and there's probably no sound card (and certainly no DOS card) that can play more than one compressed stream so only thing it can be is reducing storage requirements but in a game that has to make more than one sound at the same time it isn't really very useful, it will be a significant CPU overhead if you decompress as you go or RAM overhead if you decompress all sound load time.

Tiido wrote:

Problem is that it is not possible to mix two ADPCM sources without decompressing them first, and there's probably no sound card (and certainly no DOS card) that can play more than one compressed stream so only thing it can be is reducing storage requirements but in a game that has to make more than one sound at the same time it isn't really very useful, it will be a significant CPU overhead if you decompress as you go or RAM overhead if you decompress all sound load time.

For older machines definitely. But even a 486 can handle this transparently. Command and Conquer does this with ADPCM music and numerous sound effects simultaneously.

It does depend on a game, in stuff like Doom having 4 channels rather than 8 makes a pretty big difference in framerate on something like 486 even, and if the sounds were compressed the difference would be even bigger.

Tiido wrote:

It does depend on a game, in stuff like Doom having 4 channels rather than 8 makes a pretty big difference in framerate on something like 486 even, and if the sounds were compressed the difference would be even bigger.

True but in that case, it would be time to kick it into Turbo mode. 🤣
Although, I must say that I have never seen a difference on any 486DX past 66MHz.

But in all seriousness, I suppose that you could either decompress the ADPCM samples:
a) at load time
b) at installation
and just store them in memory thereafter. The benefit being that you could store the samples based on the output device capabilities.