xjas wrote:

Even a 286 or 386 should have no issue playing 44kHz WAVs if it can read them from disk & DMA them to the sound card fast enough. Playing a WAV file is not CPU intensive if you have DMA.

It's when you get into mixing or resampling that it starts bogging down the CPU, and if you're using something like Media Player in Windows 9x, it's doing that. Try from the DOS command line - it should be no sweat.

Tiido wrote:

I had hard time playing WAVs on a PC104 module based on AMD Elan520 (133MHz 5x86), and simply because it only had PIO HDD access. Playing over built in PCI 100Mbit network was fine though. MP3s weren't a problem on that board though sometimes there was stutter here or there when doing other things.

Well, I used an IDE VIA VT6421A busmaster card but it made no difference.

The main problem was my Asus PCI-USB2 (NEC controller) 😵
After removing it, I could play MP3 128K @ 44KHz stereo on my AMD 5x86 133@160MHz with Winamp (CPU:100%)
But playing WAV still takes 50-60% of the CPU. Anyway to stop Windows from mixing or resampling? 😕

Also possible to tweak the system to work faster?

I also noticed when I was doing benchmarking years ago that having USB card installed reduced the benchmark results. It is noted in a thread somewhere.

Enabling slow refresh is the optimal setting, but probably won't help your situation much. You need to disable the spectrum analyser and scrolling song name in Winamp. That helps A LOT!

I've been able to use a version of MPXPlay optimized specifically for the 486 to play MP3s on my 486DX4-100. I do have to disable "HQ mode" for it to work (I presume this limits it to 22kHz mono) but beyond that it seems to work fairly well, even with high-bitrate 320kbps MP3 files. The CPU is pretty much pegged at 100% nearly the entire time though. I posted a video of it on YouTube a couple years ago: https://www.youtube.com/watch?v=0c6OKiR2cXs

Barring downsampling to extremely low bitrates, I've heard a small consensus that a DX2-66 is about the lowest you could go for reliable MP3 playback, and even then it's right on the edge. For me and my unoptimized setup, the DX4-100 is just barely enough if I stay in DOS mode (I do have an MP3 player that works on it under Win95, but I literally cannot even move the mouse while it is playing or it'll skip). For reliable multitasking MP3 playback under Windows, I gotta step up to at least my Cyrix MediaGX-180 Compaq.

feipoa wrote:

I also noticed when I was doing benchmarking years ago that having USB card installed reduced the benchmark results. It is noted in a thread somewhere.

Enabling slow refresh is the optimal setting, but probably won't help your situation much. You need to disable the spectrum analyser and scrolling song name in Winamp. That helps A LOT!

Oh, I didn't know about it until I found it by myself. What's wrong with USB card? and what thread?
Yes, those features are disabled and hopefully, I can play it without any downsampling/mono but only at 160MHz 😁
I'm sure Cyrix 5x86 can play it much better if I could find one some day.

Grzyb wrote:

matze79 wrote:

you can also hookup a mp3 decoding module on the lpt port,
so you can play with any cpu that can stream the data to lpt.

There were also some internal sound cards with hardware MP3 decoding, but they never got popular, at least not in the PC,
though I reckon they were somewhat more successful in the Amiga world.

That makes me wonder - is MP3 still a thing ?
- I thought it was delcared dead since a year or two (if it's supported by Amiga, it really has to be obsolete, hasn't it ? 😉 ).
https://www.telegraph.co.uk/technology/2017/0 … 3-declare-dead/

There's jsut way too much material out there in the format so it will never die even though I would much prefer to see FLAC and OGG to be the only things you can find 🤣. It doesn't help that many tools default to MP3 or outright won't support anything but non compressed and MP3 format. Popular playback devices often don't support any other formats either, unless you can RockBox them...

RiP wrote:

I'm sure Cyrix 5x86 can play it much better if I could find one some day.

Yes, with Cyrix 5x86/100 with optimization features enabled (as in feipoa thread) I was able to play MP3s in full quality via MPXPlay. CPU usage was near 100%, though.

RiP wrote:

xjas wrote:

But playing WAV still takes 50-60% of the CPU. Anyway to stop Windows from mixing or resampling? 😕

Also possible to tweak the system to work faster?

What Soundcard do you use ?

486DX4@100 Mhz is sufficient for playing MP3 under Windoze. Used a Pentium 66 for some time. But you had to prevent the screensaer to start to prevent stutter. Probably a DOS player needs less resources and works even with DX2/66.

WinAMP probably has one of the best integer-optimized MP3 decoders available for x86.
I suppose as with most things, it depends on what you want to focus on: performance or quality?
Back in the day I switched from WinAMP to Sonique on my Pentium 133. It took way more CPU (probably because it used floating point and generally better accuracy/signal reconstruction), but the sound quality was significantly better.

I guess the real question is: how far could you optimize an MP3 integer decoder for x86? WinAMP is fast, but probably did not go 'all out' in terms of performance, but just a decent balance between quality and performance. I wonder if you could squeeze more performance out of it by using some clever approximations/tables etc, and hand-optimzing the code in assembly for a specific CPU architecture.
I think it is an interesting subject to research at some point.

At any rate, the floating point MP3 decoders will work quite well on a Pentium, but will be very poor on 486, because the Pentium FPU is just way faster. With integer decoders, the difference is much smaller.

Wait a min...
Do you mean that integer decoders can be faster than floating-point ones?

Originally, all the decoders were floating-point, and the required hardware was:

1                Pentium 486DX2-66  486DX-50  486DX-33
2MPEG-1 stereo     ok        -          -         -
3MPEG-1 downmix*   ok        ok         -         -
4MPEG-1 mono       ok        ok         ok        -
5MPEG-2 stereo     ok        ok         ok        -
6MPEG-2 downmix*   ok        ok         ok        ok
7MPEG-2 mono       ok        ok         ok        ok
8*downmix: the original stereo signal will be played back as a mono signal
9"MPEG-1" = "MPEG-1 Layer-3", i.e. sample rates 32, 44.1 or 48 kHz
10"MPEG-2" = "MPEG-2 Layer-3", i.e. sample rates 16, 22.05 or 24 kHz

(taken from l3enc272 package)

Integer decoders came much later, for more modern hardware, probably much faster than 486, only lacking FPU.
Or not?
Is there a chance to play MP3 on FPU-less processors like 386DX, 486SX, or NexGen?

Integer decoding isn't really faster, it just enabled decoding on CPUs without FPU. Mostly it was only for smartphones before armv7 became widespread.

Grzyb wrote:

Do you mean that integer decoders can be faster than floating-point ones?

Sure they can. Especially on early x86 machines. The FPU has a rather cumbersome stack-based programming model, not a general purpose register model.
Aside from that, FPU instructions have higher latency than integer ones. Especially on 486 and earlier. The Pentium FPU was much faster, but still you had about 3 cycles for simple instructions like FADD/FSUB, where the integer ADD/SUB are 1 cycle, and you had two pipelines, so you could do two integer adds/subs per cycle. So theoretically up to 6 times as fast.

Grzyb wrote:

Originally, all the decoders were floating-point

I'm pretty sure they weren't, because that wouldn't make sense.
The reference implementation of MPEG/JPEG is floating point, but I've had integer-based JPG decoders for 16-bit x86 machines, so I wouldn't see why there wouldn't be any integer-based MP3 decoders.
Even in the era of the Pentium MMX/Pentium II, there were MMX-based decoders for best performance, and MMX is integer-only.

Grzyb wrote:

Is there a chance to play MP3 on FPU-less processors like 386DX, 486SX, or NexGen?

Pretty sure there is, although probably not at 128 kbit.

jmarsh wrote:

Integer decoding isn't really faster, it just enabled decoding on CPUs without FPU.

Sure it is.
For example, on my Pentium 133, a 128 kbit MP3 took about 25-30% CPU with WinAMP (integer), but about 60-70% with Sonique (floating point).
You can't really compare that with ARM, which has a much more modern and advanced FPU (not to mention SIMD extensions like NEON).

Hard to believe it, I was under the impression that all the MP3 players heavily used FPU...
Heh, I recall playing MP3 on some Pentium 1 machine (133 MHz or so), and it was OK...
then I tried it on a WinChip C6 200 MHz (known for poor FPU), and it was much worse, hard to do anything else while playing music!

Anyway, what's the lowest FPU-less CPU to play MP3s?
Any success stories? Anybody?

Grzyb wrote:

Hard to believe it, I was under the impression that all the MP3 players heavily used FPU...

I believe with Winamp you want to look at early versions, where they used the so-called 'Nitrane' decoder.
That's the super-fast decoder which can use integer code and even MMX in later versions.
At some point they switched to the reference Fraunhofer decoder, which uses FPU.

There are also some players that use the MAD decoder (there is/was a Winamp plugin available), which is integer: https://www.underbit.com/products/mad/

It really whips the llama's ass. Baaa!

mpg123 also has an integer (aka fixed point) decoder, and actually contains i486-specific optimizations: https://mpg123.de/

This player can handle 160kbps 44KHz stereo on a 486-100 (maybe even a DX2-80) http://www.vgmpf.com/Wiki/index.php/Digital_Sound_System

The Amiga OS "amplifier" can play 64kbps 22KHz downmixed to 11KHz on a 50MHz 68030 with 68882 FPU 😉

Well, I'm waiting for somebody to do this - https://www.youtube.com/watch?v=6Jj97NXgHw4
...but without an FPU.