Does it work proper with an older (sub v2.03 type) ATX PSU ?

I presume the RAM you tried is rated for 100MHz ). Is it ?

Also, which voltage(s) did you measure (CPU input or PSU output) ?

What do the motherboard capacitors, especially those around the CPU socket and slot look like (leaks, bulging, corrosion) ? Maybe share some close-up photos .

"Does it work proper with an older (sub v2.03 type) ATX PSU ?"
Don't have any. I think you mean just try any conventional ATX power supply instead of the HDPLEX 200. Good idea.

"I presume the RAM you tried is rated for 100MHz ). Is it ?"
Yes, it's PC133 and it passed memtest86 on another 440BX running with FSB=133Mhz. I only measured the 5V rail with a multimeter since the HDPLEX seems weak in that department. For all the other rails, I only looked at what the BIOS reported. I don't see any bulging caps. Will have some more photos later.

-ECS isnt exactly a stellar brand, they actually merged with PCChips in 2005 to give you an idea what was their target market 😀
-pre coppermine board

I would check Vcore voltage with a scope, and also Vtt in light of Fixing ASUS P2B freezes on memtest86+ modulo 20 test

Pre-PCChips era ECS boards were actually decent. I think I have a P6VXA that works great. And that's with the original G-Luxon caps all around.

OK, I measured Vcore on a capacitor on the slotket with Memtest86 running and it matched what the BIOS reported, 1.55V. Also tried another power supply, but not the pre 2.03? that Horun suggested. I think that's because the older ones had more power on the 3.3 and 5V rails? Also checked for anything wrong with the capacitors, but didn't find any. Do you see anything wrong?

spacesaver wrote on 2022-11-20, 08:51:

OK, I measured Vcore on a capacitor on the slotket with Memtest86 running and it matched what the BIOS reported, 1.55V.

_with a scope_, and also Vtt. Vtt is "VCC1.5", first pin of Slot1 - A1 http://ps-2.kev009.com/eprmhtml/eprma/f706.htm I would also look at 3.3V going to the ram.
We are interested in ripple, swings and drops.

_with a scope_, and also Vtt

Heh, didn't realize how un rigorous my testing is. In mathematics, saying ∀ - something holds for all cases is hard.

I set a trigger on my DSO Quad 203 to ~90% of the regular voltage. The trigger level was set visually, so not too precise. It's the T arrow on the left side. The 2 arrow is 0V. The trigger mode is "single" - stop measuring after trigger.

I checked both VCC1.5 and the 3.3V capacitor closest to the DIMMs. But it didn't trigger even during the "block move" test that freezes. Maybe my time scale of 2us is too long and might miss a sudden dip. The capacitor on most rails is 1000uF, so I estimate the RC time constant should be in the uS range. I know the trigger works because I tested it by unplugging the probe (2nd picture)

even 9% drop on 1.5V might be bad, look at voltage in running mode without trigger. I cant tell what scale that scope is at just looking at the picture, but Im sure you read voltage correctly 😀

>2. change BIOS settings like disabling cache, increasing MD strength, spread spectrum - no difference
all ram timings to the slowest too, right?
it still did it with cpu cache disabled? 😮

in manual bios screen shows Vcore setting, can you actually bump voltage from bios? have you tried +0.15V or even +0.2V? Bumping 3.3V would be better (boards like P2B casually run at 3.5V for stability) but that looks impossible here without heavy modding.
Have a look at clock generator chip on this board, maybe something like setfsb has support and you could test other frequencies

Rasz, ignore the vertical scale. The ADC calibration is messed up and I don't have time to fix it. The 1st image was without any triggering.
I repeated the earlier voltage drop tests by setting the time scale to 0.5us and with a trigger level close to 95% of regular. Didn't trigger.

Also tried increasing voltage to 1.65 (was unstable at 1.7) and disabling cache but didn't help.

But I think I know why now. All the Pentium 3s I used are Tualatins, which use 1.25V AGTL signaling while the 440ZX chipset uses 1.5V AGTL+ signaling? I guess having the CPU receive higher voltage signals shouldn't be a problem (unless it exceeds VCC). But the chipset receiving lower level signals is bad. I don't know if raising the CPU voltage also raises the AGTL voltages.

The strange thing is I also have another 440BX system (Asus CUBX-E) without official Tualatin support, but it's able to run a pin modded SL6BY fine @ FSB=133MHz

spacesaver wrote on 2022-11-22, 08:51:

But I think I know why now. All the Pentium 3s I used are Tualatins, which use 1.25V AGTL signaling while the 440ZX chipset uses 1.5V AGTL+ signaling? I guess having the CPU receive higher voltage signals shouldn't be a problem (unless it exceeds VCC). But the chipset receiving lower level signals is bad. I don't know if raising the CPU voltage also raises the AGTL voltages.

The strange thing is I also have another 440BX system (Asus CUBX-E) without official Tualatin support, but it's able to run a pin modded SL6BY fine @ FSB=133MHz

CPU will use whatever AGTL signaling levels motherboard feeds it on Vtt pins, Vtt powers CPU output buffers. So your P3 is running on 1.5V.

> All the Pentium 3s I used are Tualatins
hmm, yeah trying coppermine P3 would be a good idea

Afaik ZX was artificially lower binned BX, intel limited max memory capacity (can be done by omitting pins when packaging) and nothing else?

Ugh, just tried a 1.1 GHz Coppermine but memtest still locks up on the block move test.

"CPU will use whatever AGTL signaling levels motherboard feeds it on Vtt pins, Vtt powers CPU output buffers. So your P3 is running on 1.5V."
Ah, I see there's separate power pins for IO (VCC 1.5) and power (VCC core) on the slot1 connector. Earlier, I was thinking the CPU output signals were some fraction of Vcore.

Yes, and Vtt is critical to stability of a system, but you already checked that above with oscilloscope (VCC1.5)?
Hmm The only options left that I can come up with are:
-find Mendocino Celeron capable of overclocking to 100Hz FSB and test that
-mod onboard integrated PGA socket for coppermine (connect two pins with a wire, pull one receptacle from the socket)

Something overheating maybe? Might be worth trying pointing a desk fan at the whole board just to make sure there's lots of air flow. I had a board that would pass all tests except block move and mod20 (not a lock up like yoursthough, just errors) on the second pass, even when testing using several different sticks of RAM. I assume those tests particularly thrash RAM contents and it turned out the RAM was overheating, specifically the chip right at the top where there was a dead spot in air flow. Fixed it by pointing a fan in the area.

mod onboard integrated PGA socket for coppermine

Wow, never thought that was possible since PPGA sounds much more different from FCPGA (Coppermine) than FCPGA with FCPGA2 (Tualatin)
Anyways, it's not worth it. I'm just going to get another microATX motherboard. The reason I was try that much on this one is because as my username suggests, I want a space saving microATX with 440BX or ZX and with AGP & ISA for the broadest compatibility. That's pretty hard to find, but I did find another one, https://theretroweb.com/motherboards/s/msi-ms-6156zx

Something overheating maybe?

I . Why would it overheat? SDRAM was from the great times when Denard scaling was still happening. By shrinking linear dimensions by 2, you got 4x more transistors, 2x faster frequency, and for the same power per area.
It's not like post 2000s where you still get 4x more transistors per area, but power density also doubles

I touched some chips and at worst, it was slightly above body temperature.