Other components of the system: Asus PVI-486SP3 motherboard, 5x86-133MHz, 16MB RAM, and an SMC 8416T ethernet controller.

Sorry to necro this, but I finally got around to doing some more testing and I thought some folks might find this interesting. Some further diagnosis:

- Repeated from above, but this is an important diagnostic criteria, so I'll repeat it. The issue ONLY affects the 16-bit DMA channel, 16-bit audio over 8-bit DMA channel works fine.
- The sound card behaves the same on a AOpen AP5VM Socket 7 board as it did on the PVI-486SP3, so we can rule out system-specific conflicts.
- I found that slightly flexing the board while playing the sound sometimes improves the audio quality. At one point after flexing the problem went away for several minutes before coming back. This points to a bad trace or broken solder joint.
- I made a recording of the 16-bit testing sound from diagnose.exe using the 8-bit DMA channel, then using the 16-bit DMA channel while flexing the board. Then I fed the WAV to ChatGPT (5.6 Sol high effort) and had it analyze it. Based on this, it strongly suspects the SD13 line. Here's the ChatGPT transcript: https://chatgpt.com/share/6a7739ce-1b04-83ea- … 4d-3ec0de46be60
- I'll do some more probing of the board out of the system as ChatGPT suggested and I'll update again if I isolate the problem.

Taking ChatGPT's advice, I traced from the ISA fingers to the pins CT1747 chip.

The routing is pretty straightforward:
- The 8 data lines on the front of the board go to 8 vias immediately above the fingers.
- On the back of the board, there are 8 short horizontal traces to reverse the order of the lines which then immediately go back to the front of the board through another series of vias.
- On the front, the traces continue directly up to the chip on the front side via short vertical traces. The data lines are all along the bottom edge of the chip, I think DA13 is pin 12.
- There weren't any obvious scratches or corrosion on any of the traces.
- The resistance measures close to 0-1 ohm all the way from finger to pin on all the data lines.
- Flexing did not produce any obvious differences in resistance, although I had the probe pressed on the pin so it could be holding it down.

At this point I decided to try reflowing the vias and pins along the bottom of the chip. This appears to have fixed the problem. Unfortunately I was too aggressive with the heat and caused the PCB to blister under the traces and solder mask. Luckily the damage appears to be cosmetic only and the board is working normally now.

Full transcript: https://chatgpt.com/share/6a774c6b-4014-83ea- … 0a-f2cc0117ec51

To be clear for those reading the transcript, I didn't intend to use 450C--I don't use my hot air station often and maybe I brushed the temp knob when I took it out of the cabinet. My brain isn't a native celsius speaker and I didn't pay close enough attention to the temp reading.

I think ChatGPT could well have been hallucinating about the specific bit that was bad when analyzing the audio. But I think it was right about it being some data bit in the upper byte and it pointed me in the right direction. I just wish I had been more careful with the temp control on my hot air station. Oh well, live and learn. The board seems to be stable for now at least during several hours of testing.