Thanks

Small updates on the bioses.

Unmodified bios 0505 and 0817 seem to be working well enough to boot up.
0817 patched with bios patcher 4.51 is also working well.
0817 patched with bios patcher 6-alpha9 hangs in the middle of writing the CPU during POST - i see the type (for example Celeron Mendocino) but it hangs before it is able to write out the speed.

All the bioses look different in the setup screen compared to the supposed "rom.by patched bios" of 0817. The layout is even slightly different. I wonder, what bios is it really based on? Is it really some other board's bios that happens to somewhat work on this one?

Form ~20 year old forum posts i read that this board should easily support all coppermines and working setups can achieve speed slightly over 1GHz. On those higher extreme cases, some say that it is necessary to use slotket that can "force" 1.8V supply to CPU. Reasoning being that those higher-end CPUs require lower voltage than the board can supply and therefore wouldn't boot.

I wonder now, what else can i check or adjust to get any coppermine working on my machine. The weirdest thing is the behavior of the board "getting tired" - when i have ran it with coppermine for 5-10min it becomes unstable and stays unstable throughout the restarts and so until it can "cool down" over night or so.
An unfortunate addition is also that the Celeron 300A@450 has also started to hang after 10min of looping 3dMark2001 low-detail car demo. 🙁

What is going on?

UPDATE: Some more testing and one weird thing i've figured is that my 866MHz coppermine, which should request 1,75V voltage, is getting either 1,45V or 1,5V from the board. (it's about 1,479V or so, pretty much in the middle there.)
I've "beeped through" all VID pins from slotket to the voltage controller chip and all 4 pins show connection...
Some VID0-3 pin magic issue? 😮 Unfortunately none of my slotkets allow to manually set voltage and the board's bios doesn't allow it either...

A little update.

I dug up my SE440BX-2 and tried my slotket with 866MHz Pentium 3 and it is happily looping the 3DMark2001 low-detail car demo. (Running at 6.5x100MHz)
I really dislike that board because of the bios that is very non-hackable (Phoenix 4.0), has almost no options for any settings and the fact that newer BIOSes block newer CPUs even when the hardware on board really supports it. (I checked the voltage selector chip and whatnot; Currently running P13 bios, while the latest is P17)

So i can confirm that i have "stable hardware" around the CT-6BTM0 board and i need to somehow solve the issues that this board itself has...

GigAHerZ wrote on 2024-06-27, 08:51:

So i can confirm that i have "stable hardware" around the CT-6BTM0 board and i need to somehow solve the issues that this board itself has...

FWIW I have two slotkets and a few S370 CPUs (as well as Slot1 CPUs, everything was gifted to me except a mobo) and one of the slotkets works pretty much perfectly in Slot1 mobo I have, the other craps out eventually in Doom and Quake demo loops with the same CPU. This might be a slotket issue (the unstable one doesn't have resistor packs for signal termination) or a mobo issue, or a combination of both. I'm not much into P2/P3 so I don't want to buy another Slot1 BX mobo just to find out which part is the problem. And frankly even that wouldn't be definitive, after all I could just have two somehow subpar mobos on my hands.

I know slotkets are a thing but as far as I'm concerned if you want to run Socket 370 CPU and not worry about stability then you need Socket 370 mobo. Other combinations might work, and sure do for many people, but are not guaranteed to work. I've recently found a very cheap S370 mobo so I bought it even though it might be dead. We'll see. If I can get it working I'd have a platform for some of the CPUs and could just sell or stash the suspect slotket.

@Deunan, slotket is just one option i have here. The Chaintech board is unstable with SLOT1 P3 CPUs, too.

GigAHerZ wrote on 2024-06-27, 11:11:

The Chaintech board is unstable with SLOT1 P3 CPUs, too.

But these are all 133MHz P3s, right? Could be something about them that makes the mobo unhappy, even when running at reduced FSB speed. You'd need something like 400 or 450 MHz P2 chip to verify if this mobo can handle 100MHz FSB at higher clock rates. And even that is still below 30W TDP, while the P3 chips at 2V draw more power. Yours might be the 1.5V variants (you didn't specify) but the slotket power regulation might not be dealing well with that, or the CPU is causing too much noise and something just can't handle it. Could be anything - mobo routing, tired caps, PSU with a bit more ripple, etc.

Deunan wrote on 2024-06-27, 13:13:

GigAHerZ wrote on 2024-06-27, 11:11:

The Chaintech board is unstable with SLOT1 P3 CPUs, too.

But these are all 133MHz P3s, right?

Nope. I have some 100MHz FSB P3-level CPU's too. I even have some 66MHz FSB Coppermine-128 CPUs.
I have tried to be specific on every CPU in this thread. Most information is in this post: Re: Fixing CT-6BTM0 motherboard - C0, C1 and then ambulance alarm
I have no CPUs with so low voltage as 1.5. Most are around 1.7, some are 1.65 and i do have slot1 P2 and P3 with 2V requirement.

GigAHerZ wrote on 2024-06-19, 20:40:

UPDATE: Some more testing and one weird thing i've figured is that my 866MHz coppermine, which should request 1,75V voltage, is getting either 1,45V or 1,5V from the board. (it's about 1,479V or so, pretty much in the middle there.)
I've "beeped through" all VID pins from slotket to the voltage controller chip and all 4 pins show connection...
Some VID0-3 pin magic issue? 😮 Unfortunately none of my slotkets allow to manually set voltage and the board's bios doesn't allow it either...

On this one, i confirmed it that it is not an issue. I was measuring wrong point. There is a mosfet with 1.5V and also with CPU voltage. I picked a wrong one. So the voltage should not be an issue.
... but i did order a slotket with manual voltage control already. 😀

GigAHerZ wrote on 2024-06-27, 15:28:

Nope. I have some 100MHz FSB P3-level CPU's too. I even have some 66MHz FSB Coppermine-128 CPUs.

Hm. Correct me if I got it wrong or missed something again - the best 100MHz CPU that still works fine is P2 350MHz? If so that's only 21.5W TDP according to wiki. Back then TDP actually meant something, perhaps not the total power but it would not be a random attribute for PR reasons. The higher clocked CPUs would draw more, and if your problems are most easily triggered by running something graphics heavy then I would still suspect power delivery over the mobo. Could be the issue is not CPU but the GPU HW that crashes (or the mobo chipset), it's just that with faster CPU there is more power draw and that's required to trigger it.

BTW you mentioned oscilloscope - you are right to suspect that 250mV ripple on the 3V3 line could be your measurement issue and not actual ripple. It's always good to have some idea what to expect, and if the measurement is considerably off to doubt the instrument rather then the circuit. It gets easier with experience.

Most likely you've picked wrong GND point. The higher the frequency the more important it is to pick a reference point as close to your measument point as possible, otherwise you gather all the noise from other chips - common-mode noise as well - into your measurement. So typically you want that ground clip attached to a nearby capacitor that filters that voltage rail, and keep the wire short. DC coupling is preferred unless you want to measure really small AC signals on top of high DC voltage that could overload the input, but in this case your scope should have enough resolution to resolve ripple down to mVs or so even on 3V3 line. For 12V and really small ripple you can start considering AC coupling. As general rule AC coupling will pick up more noise, including noise radiated as EM waves. If your scope has the option to limit the input BW to 20MHz this is the very use case for it - you are not expecting to see such high frequency on power rails, so might as well try to filter it out.
1:1 or 10:1 doesn't matter much for low-impedance sources like power rails, and at low frequency. You want to stick to 10:1 to avoid loading the circuit with the scope probe, but obviously this will attenuate the signal 10x. Expensive scopes and probes will talk to each other and auto-adjust the display but for typical passive probes you need to set the scope to match the probe settings, otherwise the display might be 10x off vs what you measure. The waveform will be the same though.

Deunan wrote on 2024-06-27, 17:51:

GigAHerZ wrote on 2024-06-27, 15:28:

Nope. I have some 100MHz FSB P3-level CPU's too. I even have some 66MHz FSB Coppermine-128 CPUs.

Hm. Correct me if I got it wrong or missed something again - the best 100MHz CPU that still works fine is P2 350MHz? If so that's only 21.5W TDP according to wiki. Back then TDP actually meant something, perhaps not the total power but it would not be a random attribute for PR reasons. The higher clocked CPUs would draw more, and if your problems are most easily triggered by running something graphics heavy then I would still suspect power delivery over the mobo. Could be the issue is not CPU but the GPU HW that crashes (or the mobo chipset), it's just that with faster CPU there is more power draw and that's required to trigger it.

Yep, "P2 class" CPUs have shown better stability. But it has been a bit confusing, as i was able to observe some hangs event with my Celeron 300A recently.
Some efficient Coppermines with downclock (because of 100 instead of 133 MHz FSB) should lower the TDP quite a bit, shouldn't it? And for example my 66MHz FSB Coppermine-128 (Celeron 766 with 66MHz FSB) is rated at 20W TDP.

Deunan wrote on 2024-06-27, 17:51:

BTW you mentioned oscilloscope - you are right to suspect that 250mV ripple on the 3V3 line could be your measurement issue and not actual ripple. It's always good to have some idea what to expect, and if the measurement is considerably off to doubt the instrument rather then the circuit. It gets easier with experience.

Most likely you've picked wrong GND point. The higher the frequency the more important it is to pick a reference point as close to your measument point as possible, otherwise you gather all the noise from other chips - common-mode noise as well - into your measurement. So typically you want that ground clip attached to a nearby capacitor that filters that voltage rail, and keep the wire short. DC coupling is preferred unless you want to measure really small AC signals on top of high DC voltage that could overload the input, but in this case your scope should have enough resolution to resolve ripple down to mVs or so even on 3V3 line. For 12V and really small ripple you can start considering AC coupling. As general rule AC coupling will pick up more noise, including noise radiated as EM waves. If your scope has the option to limit the input BW to 20MHz this is the very use case for it - you are not expecting to see such high frequency on power rails, so might as well try to filter it out.
1:1 or 10:1 doesn't matter much for low-impedance sources like power rails, and at low frequency. You want to stick to 10:1 to avoid loading the circuit with the scope probe, but obviously this will attenuate the signal 10x. Expensive scopes and probes will talk to each other and auto-adjust the display but for typical passive probes you need to set the scope to match the probe settings, otherwise the display might be 10x off vs what you measure. The waveform will be the same though.

Thank you for those pointers!
I'll try to find some time soon again to see what i could measure with ground connected to some nearby ground reference. I previously used some LPT or COM port shield as ground reference.
I'll have to check for the bandwidth limitation. Supposedly it is 110Mhz unit, but as it is a cheap chinese clone, reality might be closer to 50Mhz. (Hanmatek DOS1104 - Love the name as it has "DOS" in it! Perfect for retro stuff 😁 )

GigAHerZ wrote on 2024-06-03, 18:01:

So here's a PSA! The cheapest ATX extension cables from china are using below-standard wire widths! 18AWG should be standard so make sure your extension cable is at least that.

I got another extension. This time clearly marked also as 18AWG. But as an article, it's different as it was really meant for those cryptominer guys to start multiple PSUs with single motherboard.
It's as bad as previous. With no extensions, 5V is rock solid at 5V. With extension, the voltage drops to 4.65V measured from motherboard ATX socket.

Damn...

two in a row suggests the problem is with your oxidized connectors and not the cable

rasz_pl wrote on 2024-07-05, 22:02:

two in a row suggests the problem is with your oxidized connectors and not the cable

Connectors cause the wires themselves become warm? And connectors directly connected do not cause voltage drop?
Dude... 😁 It's about chinesium cables and the fact that nobody today expects there to have high-current 5V capability on ATX supply.