bertrammatrix wrote on 2025-05-29, 04:58:

I'd ALWAYS use the ATX if there is that option, dont even bother with the AT, HOWEVER, make sure that your power supply actually has the -12v wire, later supplies often lack this and it can make weird stuff happen

I think you've meant -5V? -12V is pretty much on all half modern PSUs.

I realize this schematic is probably not an exact match, but the main arrangement of power components looks the same as your board: 3 large semiconductors, 2 inductors, bulk capacitors. So most of the power (Vcore) is handled by L1 L2 D1 and Q2. These beefy components are mostly lined up in a row along the outer edge of your board. Q1 handles the I/O supply, which is the large FET closer to the socket in your pictures. I would guess that this is where your 3.5v is coming from.

You can see that if the detection pin is not grounded (inside the CPU) then the 5v input to R8 will pull up the I/O feedback pin (through D2) and disable that supply. Then the core voltage (or ALL voltage on a single voltage CPU) is determined by the R4/R5 resistors to the second feedback pin. On your board, R5 is probably a parallel arrangement with whatever jumper you select from an array that can optionally raise the voltage from a baseline.

If the detection pin is floating (not connected to anything) then its exact voltage is related to the values of R6, R7, R8 and the I/O output voltage. I estimate you would measure somewhere between 2.5v and 5v; so 0.6v indicates a problem.

I've also seen cases where the detection pin will shut down the Vcore supply instead, and everything runs on the much weaker and non-adjustable I/O voltage source. You should be able to find two different feedback pins on the control chip, and follow the detection pin to one of these through a small diode and resistor network.

If your control chip is really only 5v, then some of the other small transistors may be the gate drive for the large ones. So you should be able to find 12v there. Be careful probing gate or control signals on a live board. You can inadvertently turn on a FET and put a voltage spike through the whole board.

If you can't figure out what the VRM is doing, it looks like you could remove the L2 inductor and connect your own separate VRM to the CPU side pin, and supply an independent core voltage there. You may need to add a pull down resistor to keep the mostly floating Vcore in check. If you go that route, this thread may help:

www.vogons.org/viewtopic.php?t=86611

One of the socket-VRM devices like the Evergreen Spectra would also be an easy way to do this, or at least test whether the rest of the board is working properly before going further.

nickles rust wrote on 2025-05-29, 12:42:

If your control chip is really only 5v, then some of the other small transistors may be the gate drive for the large ones. So you should be able to find 12v there. Be careful probing gate or control signals on a live board. You can inadvertently turn on a FET and put a voltage spike through the whole board.

Oddly, I found 12v on an "72" SOT-3 (2n2222?) which is close to the schottky diode on the source pin.
As for now, I'll see what I can figure out. As a last resort I'll need to build my own VRM board.

Edit:
Found the culprit for the 0.6v on Vccdet. it's kinda an "RIP" marked SOT-323 transistor at fault. Gate goes to Vccdet, drain goes to Vout of an Vref SOT-325 by an resistor. While source goes directly to GND. Without it, Vccdet goes to 5v, the drain of that transistor has 1.6v, and gate 5v.

Any ideas?

There are all kinds of other VRM controllers on socket 7 boards that you can look at to try to understand what's happening. There's even this thing that forces a shunt regulator to oscillate.

kotel wrote on 2025-05-29, 04:40:

Both Vcore pins on my AR7010 are connected together (0 ohms).

It's a really strange-looking layout then. I don't think it means anything (that trace length wouldn't do much to add inductance or delay at 100-200kHz).

kotel wrote on 2025-05-29, 04:40:

As for the FET above Q11, drain goes to Vcore, gate to 12v and source to Vi/o

That would mean that it's the FET I mentioned that bridges the core and I/O supplies when VCC2DET is not grounded. 12V on the gate is what I would expect if VCC2DET is not grounded.

It also means that I have no idea where the I/O voltage is generated. It would totally make sense to use that FET as a linear regulator like the US3033, but that isn't what is happening here. The equivalent part in the schematic in the post I linked above is M3.

kotel wrote on 2025-05-29, 04:40:

For the ATX part, the board does the same thing whether I use an AT or ATX PSU. But, according to my POST card, the 3.3v is missing when I use an AT PSU, meaning there's no onboard 3.3v generation.

That just means that PCI 3.3V power is supplied from the ATX connector only. This is a super socket 7 motherboard, so it would be pretty stupid to put an unusable AT power connector on it.

kotel wrote on 2025-05-29, 15:43:

Found the culprit for the 0.6v on Vccdet. it's kinda an "RIP" marked SOT-323 transistor at fault. Gate goes to Vccdet, drain goes to Vout of an Vref SOT-325 by an resistor. While source goes directly to GND. Without it, Vccdet goes to 5v, the drain of that transistor has 1.6v, and gate 5v.

0.6V would be correct if they used an NPN transistor instead of a FET. I looked around, and it looks like Rohm-manufactured 2N2222s are marked R1P.
https://www.s-manuals.com/smd/r1

What's the Vref? That sounds interesting.

lti wrote on 2025-05-31, 01:41:

What's the Vref? That sounds interesting.

Markings on it are "D00B" (zero's can be "O"s) in SOT-235 case. Seems to be an LM3411M5-3.3. Although, I don't get why I do not have 3.3v on output even without the 2n2222 transistor....

What voltage do you see on the input pin? It might be part of the feedback to the mystery controller (in the same part of the circuit as the voltage selection jumpers). It's weird to ground the output, so there must be a resistor in series with the input to limit the current.

I still don't see why they didn't just use that big FET as a pass transistor for a 3.3V regulator. It could be something like figure 28 in the datasheet for the LM3411 (page 20), but modified for higher current (and a FET instead of the D41D5 PNP transistor).
https://www.ti.com/lit/ds/symlink/lm3411.pdf

lti wrote on 2025-05-31, 15:50:

What voltage do you see on the input pin? It might be part of the feedback to the mystery controller (in the same part of the circuit as the voltage selection jumpers). It's weird to ground the output, so there must be a resistor in series with the input to limit the current.

I still don't see why they didn't just use that big FET as a pass transistor for a 3.3V regulator. It could be something like figure 28 in the datasheet for the LM3411 (page 20), but modified for higher current (and a FET instead of the D41D5 PNP transistor).
https://www.ti.com/lit/ds/symlink/lm3411.pdf

Measurements on Vref component:
Vout: 2.63v (goes through an resistor and into that RIP transistor)
Vin: Vcore voltage (3.50-3.52v depending on how long did the board run)
Comp: changes when I set the Vcore jumpers (1.4v while vcore is set to 2.1v, 2.5v when vcore is set to 3.2v)

As for that design, who knows. It's epox.

The COMP voltage changing means that it is part of the feedback for the VCC2 (Vcore) supply. The output should eventually connect to the AR7010 after going through some other parts (probably just some resistors). Drawing the circuit as you go (including all connections to VCC2DET) might help.

Were your measurements with the R1P transistor reinstalled? Did you try measuring with VCC2DET grounded to see if anything changes?

I measured with the R1P transistor. Vccdet is still 0.6v, so low chances it will actually change anything. Wondering if this Vref component is somehow bad.
Comp pin goes to an via which then goes inside the board (this PCB seems to be triple layer).

lti wrote on 2025-06-01, 19:15:

The COMP voltage changing means that it is part of the feedback for the VCC2 (Vcore) supply. The output should eventually connect to the AR7010 after going through some other parts (probably just some resistors).

COMP pin connects to the Vcore jumper block (on the side where the text is). There might be an second trace going in the middle of the board, but I haven't seen any low-mid resistance connection to the AR7010.

I meant that the OUT pin would connect to the AR7010 (after the resistor and transistor you found). I would also expect a connection from the R1P transistor to Q9.

Did you figure this out? I noticed that Scrap Computing also had difficulty trying to test a socket 7 board without a CPU installed:

/watch?v=MigVZLc_Mgw

/watch?v=-WO0Vd2BWVc

nickles rust wrote on 2025-06-09, 17:55:

Did you figure this out? I noticed that Scrap Computing also had difficulty trying to test a socket 7 board without a CPU installed:

/watch?v=MigVZLc_Mgw

/watch?v=-WO0Vd2BWVc

Nope, it's currently sitting on the shelf waiting to be repaired. I still have some simpler problems I need to fix asap.

As for the videos (they are on youtube if anybody is wondering, two parts about a butchered ASUS P5A), they gave me courage to try again. Although, I did already insert my only CPU at the begining and it didn't do anything....

Now, just for my info, could somebody here measure the resistance of Vcore and Vio to GND on their working K6-2's, please?

Okay, I yolo'ed it and put the CPU and Vcore set to 2.1v.
I did get an C1 post code (no RAM currently), but the Vcore was still 3.52v which is too dangerous to run an 2.2v k6-2 at.....
The OUT pin doesn't connect anywhere to the AR7010 chip.
Besides, this whole VRM is just pure confusion. Why bother with an 5v VRM just to drive some small FETs to switch the bigger 12v ones? This is just overcomplicated.

Any ideas?

Remote troubleshooting like this is difficult, especially when what appears to be the controller has no known public datasheet and the same part number as an extremely common Atheros USB WiFi chip. It still feels like it isn't detecting the multi-voltage CPU correctly, but it's hard to say without having the board in front of me.

The VRM controller needing external 12V gate drive is strange since there are lots of other socket 7 VRM controllers that can connect directly to the FET. It isn't as bad as that Anigma 586, but there are much simpler designs like the part from the later revision. Also, FIC liked to use the SC1101, which is a nice and simple part with built-in current sensing.

Tried an older k6-2 (this time an known good k6-2 200 ALR) and the board still generates the stupid 3.5v Vcore regardless of jumpers. Although the board does execute POST codes with any CPU (with no RAM it goes to C1, to be expected). So at the very least the board works but the VRM doesn't. Will keep an eye out for the very early ones which didn't have the IO voltage.
Recently I also got an P5T30-4B board which also had 3.3v Vcore regardless of jumpers without CPU, but with it inside the voltages are fine.

In the meantime, does anybody have any ideas before I go modding the VRM to take my own one? I did make sure I wasn't measuring the Vio rail.

Found the issue!

Turns out, I was wrong. The coil is for Vio, not Vcore. The VRM is working properly and responds to Vcore jumpers. The Vcore MOSFET is the one above the schottky diode!
Sure, Vio is a bit high (should be 3.3v) but at least we know this was just an issue with understanding of the VRM.
But now, this doesn't add up. Vcc2 is Vcore while Vcc3 is Vio. Vcc2 is the middle pin on the cut corner of the CPU. It is connected to 3.5v while Vcc3 is connected to the MOSFET which responds accordingly to the set jumper.
Measured it wrong. Vcc2 was NOT connected to the coil but directly to the CPU pins! Vcc3 goes through the coil into the VRM chip and back out of it.
At least we've figured out the pinout for this bizzare chip.

Will do further testing to confirm whether VRM is fully working.

Okay, the VRM is bad. Not only Vio is stuck at 3.54v, the MOSFETs get hot when idle (70-80ish C by finger).
All caps were replaced with Nichicons. The caps that get warm on their own (20-30C) are on the 5v in rail (the one that feeds the whole VRM). Even the AR7010 gets a bit hot while idle.
Coils were replaced with those who have more coil turns and they did help, but not enough to keep these temperatures in spec.
This board is fishy. When I got it the MOSFETs unsoldered themselves, so maybe they are bad but my LCR T3 tester said they were fine?
With SQM50034E MOSFETs the same thing happens. Although the one that gets hot is the Vcore one. I don't know whether the VRM on these revisions are bad or what, but this isn't the first case of those MOSFETs desoldering themselves.

It's starting to make sense now. Vcore is generated by a linear regulator supplied by Vio. That's a really stupid design.

Vio might be that high intentionally. Asus did that on the P5A (the minimum Vio on the P5A was 3.5V, and the P5A-B even has jumper settings up to 4.0V). It could be for overclocking stability (that was probably Asus' thought) or because the Vcore linear regulator didn't work at 3.3V. The FET closest to the socket (the one for Vcore) getting that hot is probably normal since it's a linear regulator. How hot does the other FET get?

I also found a Japanese page about this board where someone replaced those two FETs. You can run this through your translator of choice (Google Translate is broken right now - it defaults to German-to-English, and the auto-detect thinks this is Simplified Chinese).
https://www.asahi-net.or.jp/~qr5n-kt/pc/topic20020712.htm

So, after I resoldered the FETs somehow all the issues went away? Odd. At least the board works fine now.