I'm running at 160MHz stable at fastest RAM & L2 cache timings now 😀
I measured and replaced the LDO by Linfinity LX8383A-00.
I set Vcore @3,7V but real measured average voltage is 3,45V. There's not negligible Vdop on the power path...

scopes here:
http://rayer.g6.cz/hardware/retropc2.htm#EXP4044_VRM2

With no load, you measure 3.7 V from the regulator, but with load, you measure 3.45 V? That voltage drop is higher than I'd have expected. LX8383A-00 is a 7.5 A regulator, so you are not even close to exceeding this.

Noise of 500 mV seems a bit high. Where are you measuring? How about measuring Vpp with a low inductance probe tip on a) the VRM, b) a capacitor on the interposer's PCB?

I wondered too. Seems that PCB layout is not much optimal, also Vcore goes via some jumper (5V fixed/VRM) and probably thin traces until it reach the socket...
I measured noise with GND of probe clipped to one jumper close to socket and tip of the probe was touching one of blocking capacitor on top side of interposer.
I know that noise changes on different places but it was intended mostly for comparison between LDOs and there was not much difference...

BTW how about the VLBus wait states? This MB has only one jumper that is markded as VLB <=33MHz / >33MHz. I tried both positions and VGA speed was not affected. What only makes the difference is if I change the ROM image by 0WS/1WS option (there are available 2 ROMs for S3 765).

Attached heatsink with homemade spring for Socket 3

Good job, RayeR. I have two questions.

Finally what voltage regulator you've soldered on this motherboard for using the 3.3 V CPUs?

Will you share what material it is and how you made the homemade bracket for the heatsink? I'm considering a similar DIY heatsink bracket for socket 3 CPU using some kind of steel wire, but I'm still at the ideas stage.

As I wrote above, I used LX8383A-00 but LM1084-ADJ would be fine too, just couldn't find one.
I modified R63 to 220 ohm to get 3.7V in theory, 3.45V real under load.

I just took some ~1mm dia steel wite and bent it by tongs to somethinkg like "M" shape, where the middle of M pushes on the center of the heatsink and edges of M are hooked by plastic noses on the socket sides.

analog_programmer wrote on 2024-11-22, 17:16:

Good job, RayeR. I have two questions.

Finally what voltage regulator you've soldered on this motherboard for using the 3.3 V CPUs?

I'll share as well, although, keep in mind that I only upgraded the machine to a standard DX4 CPU. My motherboard didn't have the voltage regulator either, but it did have the resistors populated already so that the regulator could just be "dropped" in, so to speak.

I put in an ordinary LM1084-ADJ; the value for R63 is 249Ω, and R65 is 432Ω. That results in 3.41V on the Vcore pin.

The regulator was placed in Q1A, with the pins across pads 1, 2, and 4 as shown in the picture below (and on the first page of this thread). The heat sink is small, but it was all I had for the moment.

Interesting, my MB had R63 - 270ohm, so the voltage was lower. It seems they did some development steps and maybe they also found that it's not 100% stable at 3,3V and decided to rise Vcore a bit...

If you direct some air flow from CPU fan or PSU fan over LDO heatsink it could be OK...

Well, I wouldn't say the board is entirely stable. Prior to the mod to add the 3.3V rail to the processor socket, I had a DX-33 processor in it. Sometimes, it would boot normally, and other times, I could start it and while it would reach the POST screen, it would never start the memory count and it would just beep with a constant tone. It was always a toss-up as to whether it would start. Even post-modification, it still does it sometimes.

Once it did, it was fine, and no amount of stress testing would iron out what the issue was.

However, this board gets even weirder, because while I could never quite pin down why it would just flake out during the POST process, I did notice that pin 39 of the W83c42 (the keyboard controller) was clipped. As in, someone prior went in and actually cut pin 39 off of the chip. There was enough there that I could solder a tiny lead fragment to, and when I put it back in the socket, the system would make it to the initial POST screen and register 640K of memory and just ... stop. No more counting, no more nothing, just a dead stop. I pulled pin 39 back out and all of a sudden it was fine.

I am baffled by why this board behaves the way it does. Other than that, it's a solid machine.

Hm, mine works normal with pin 39. Maybe someone replaced KBC from other MB?
Datasheet of i8042 shows that pin 39 is TEST1 input:
TEST INPUTS: Input pins which can be directly tested using conditional branch instructions.
FREQUENCY REFERENCE: TEST 1 (T1) functions as the event timer input (under software control).

Interesting. Do those 3, 4 or 5 through-holes match the pinout of some suitable 3.3 V regulator? I can't find any picture of this board with factory installed VRM.

I asked for it at beginning of thread but seems nobody have it. But why bother when common LM1084-ADJ easily fits in...

I got a couple of "used" W83C42s and popped one in last night and it booted right up. Maybe something nuked the test input on the old chip to the point where it interfered with startup.

Even so, it still does the same thing every now and then -- I will start the machine and it will just completely freak out and just give me a continuous tone, and then I'll hit the reset switch and it'll be fine. It makes me think there's damage to the board somewhere that I can't see. The board does have a soldered-in lithium cell, but those usually don't have any problems with leakage, so I doubt it's battery damage. It just confuses me like crazy.

I also double-checked the heat on the regulator, and it's getting pretty toasty, which is making me think that I may be getting brownouts on the 3.3V line. I beefed up the heatsink, and we'll see if that helps. Worst comes to worst, I'll just double up on the heat sinks to try and give it somewhere to dissipate.

RayeR wrote on 2024-11-28, 01:39:

I asked for it at beginning of thread but seems nobody have it. But why bother when common LM1084-ADJ easily fits in...

You did find the one that fits in the 4-pin footprint, though. Trouble is, it appears to be a part no one makes anymore.

Oddly, I have the 9V version of it.

Regardless, an LM1084-ADJ works by just popping it into the 5-pin footprint (after a little creative bending of leads).

This sounds like your MB has some bad solder joint or damaged via on PCB that cause random disconnection and thus failure. It's very hard/nearly impossible to find it. It probably go worse during time.

One friend brought me some abit 486 PCI board to have a look on. He told that after some random period of time it hangs and after reset didn't boot, hanged at POST code 09h. He hadn't other components to try. I tried with different CPU, VGA and checked RAM and didn't find a problem. For me it runs pretty stable. I returned it to him and he told that hang appeared again. He also tried other PSU...

I decided it would be quicker and easier for me to just buy some cheap interposer boards from China and swap the CPU. The Intel CPU's will be removed.

Sure, how did it cost? I didn't find this, just old 486SX on interposer but it has lower pin count (same as on the MB).

RayeR wrote on 2024-12-14, 14:29:

Sure, how did it cost?.

Less than the cost of parts to assemble one. This type of stuff sold within China is cheap, provided you don't start asking questions.

The broker still has more to liquidate, so I don't want to reveal my exact price. Maybe he wants to up his price, I don't know.

RayeR wrote on 2024-12-14, 14:29:

I didn't find this, just old 486SX on interposer but it has lower pin count (same as on the MB).

You normally won't find these outside of China, unless you have someone on the inside, selling out. However, 10 years ago, these interposers were also sold for cheap on eBay from Chinese sellers.

I plan to use the interposers for Am5x86-133 B1 chips (gunning for 200 MHz), Cx5x86-120 (hoping for S1R3 @133 mhz), and IBM 5x86c (133 mhz backups.

Well, if you can reach such deep sources... can you also get e.g. cheap Tillamook on interposer? The best 300MHz version? Sphere was hunting for it. Or AMD K6-2+ for cache mod?

In my case I paid cca 4.5$ for PCB, 10$ the CPU on Ali, 3-4$ some cheap heades. Maybe for 20$ I could just buy ceramic CPU but my plan was to solder it directly on board but I failed to count pins before the order and nobody warned me before...

I haven't looked recently for a Chinese Tiger Tillamook MMX, but they would have been common 10-15 years ago. Unfortunately, I don't own one, but it is on my list.