I've been trying to kill the high-pitched whine/hiss in the Rev. B power supply, correlated with CPU/GPU load, for a while now and have not been successful. In addition to the recap, I tried swapping the Rev. E fan control board into the Rev. B. It works fine, but no noise improvement.
I thought I'd found a really good lead here:
https://www.badcaps.net/forum/troub...14027-acbel-9841-problems?p=351515#post351515
These are the PWM daughterboards from the Rev. E supply. The date codes on them are weeks 48 and 49 of 2001. Electrolytic caps are the usual suspects, Lelon and Teapo:
Here are the same boards from the Rev. B supply. Date codes are weeks 22 and 23 of 2001. I haven't spotted any differences between these and the Rev. E boards except the suppliers of the 3842 chips:
On the main power supply board, the small board is labeled PCB3 and the larger board is labeled PCB4. It's clear that in both cases, the smaller board has been running hot for a long time. The board and green jacket on the Lelon capacitor are both discolored. Seems reasonable to speculate that the Rev. E drove an ADC display for a long time, and the Rev. B didn't, based on the heat discoloration on the larger boards.
The electrolytic caps are:
| RefDes | Capacitance | Voltage | Width (mm) | Height (mm) |
|---|
| C303, C409 | 2.2uF | 50 | 5 | 12 |
| C309 | 0.47uF | 50 | 5 | 12 |
| C403 | 10uF | 50 | 5 | 12 |
Both of the Rev. B's 2.2uF caps were fully cooked. No measurable capacitance. The Rev. E's were hanging on at 2.2uF and 1.8uF, but its 0.47uF was cooked. All four of these should be a high priority to replace, given how hard they seem to be working in this design.
I had the 10uF and 0.47uF parts on hand, so I replaced those with new parts. I didn't have any 2.2uF caps, so I tried the tired but not quite dead Lelons from Rev. E on the Rev. B board. No improvement in noise. I have some new 3.3s, I should just throw those on. If the circuit still worked with failed caps, it can't be that picky...
Based on the BadCaps post linked above, I checked all the MLCCs on all 4 boards. I removed them individually and measured them out of circuit. All of them measured sane values, matching between boards. No evidence of them "going out of tolerance" unless Andy3665 meant that they drift excessively when hot? Could be the case with crappy dielectric. I don't think that's related to my noise problem, though; it's evident almost immediately after power on.
Here's a table of the MLCCs, just in case it's useful to anyone:
| RefDes | Capacitance | Size |
|---|
| C304, C307, C404 | 1nF (1000pF) | 0805 |
| C301, C407 | 100nF (0.1uF) | 0805 |
| C302, C402 | 10nF (0.01uF) | 1206 |
| C305, C405 | 680pF | 0805 |
| C308, C408 | 220pF | 0805 |
| C406 | 1nF (1000pF) | 1206 |
| C407? Bodge on PCB4 | 1nF (1000pF) | Through-hole |
There must have been a good reason to use a through-hole (single layer?) 1nF ceramic capacitor bodged onto the pins and glued down instead of just populating an MLCC at C407, since they kept it through multiple revisions of the main supply.
Will tackle the psu on the weekend.
