Whether you can overclock a board 33% beyond spec is a bit of a lottery, even with a high-quality board with late revision of a chipset famous for OC potential. I have a pile of MSI MS-6168 rev2 boards (late BX with onboard Voodoo3) and two of them reliably do 133MHz and two others utterly fail over about 112MHz.

As for the caps -if they are original, 24 year-old caps from the plague era are pretty suspect. You can check how suspect by measuring CPU core voltage under load. Is the board delivering the required voltage to the CPU? If so, I'd not primarily suspect the caps. If not, change the buggers.

My primary 440BX board won't run stable above 127MHz FSB. At some point in the future, I intend to modify the resistor array to juice the VIO regulator to 3.45V to see if that does the trick, but that's more of a curiosity.

Capacitors are very unlikely to be the barrier.

dulu wrote on 2024-08-02, 20:34:

3. tried on 3 PC133 ram sticks

He-he, dude, its BX. 🙁
Welcome to club

There is no 133 MHz option in the chipset registers, there are only 66 and 100
Thus, functionally this is not PC133 memory, it is PC100 memory overclocked to 133 MHz.
Any PC133 from NCP, PQI, AM (even Samsung) most likely will not work on the 133 bus.
Or you need to have a cargo container of such memory and select sticks.
Hynix - 50/50, more likely yes than no.
There will definitely be Infineon and Micron.

To be sure, you should find a stick with a clear indication of "PC133 CL2" from a decent brand for 64-128MB.
And try to start only with this memory stick

something like this
At the same time, NCP memory was tested on 140FSB@CL2 on the CUSL2-c (i815) motherboard without any issues.
On the BX it did not start on a bus greater than 100 MHz.

shevalier wrote on 2024-08-03, 06:47:

Any PC133 from NCP, PQI, AM (even Samsung) most likely will not work on the 133 bus.

Not entirely true. I have an ABIT BE6-II (440BX), SL5DV P3-1000 133FSB slotket and a PQi 128MB CL3 stick, always worked and recognized fully.
I could not run CL2 sticks on it however.

PcBytes wrote on 2024-08-03, 11:43:

shevalier wrote on 2024-08-03, 06:47:

Any PC133 from NCP, PQI, AM (even Samsung) most likely will not work on the 133 bus.

Not entirely true. I have an ABIT BE6-II (440BX), SL5DV P3-1000 133FSB slotket and a PQi 128MB CL3 stick, always worked and recognized fully.
I could not run CL2 sticks on it however.

Lucky.
A good examplar of memory and motherboard.
I saw a video where started 4 completely different memory sticks together.
How many attempts to choose a memory for this - naturally remained behind the scenes.

The CL2 module is needed to guarantee that CL3 runs at 133 MHz on the 440BX.
CL2 is an indicator of memory "quality" for 440 chipsets on a 133 MHz bus.
So as not to guess what the problem is - in the memory, motherboard or power supply.

Yeah, I have a Soyo SY-6BA+IV that I also tried just to make sure the CL2 sticks weren't bad.
On that one it's as normal - CL2 works, CL3 doesn't (or the very few that do are at half capacity.)

Ok, so the ram that worked on i815 with fsb higher than 133 does not work on 440bx even with fsb133. I did not expect that. I tested the Ram on Athlon / VIA KT133, with a 4/3 divider, so I thought that 133 was assured. I will check other sticks of ram that i have.
I'm currently testing my PQI cube on my 440bx - rated at 133/CL3, running at 112/CL2 rock stable in prime and quake 2 loop

smtkr wrote on 2024-08-03, 00:05:

My primary 440BX board won't run stable above 127MHz FSB. At some point in the future, I intend to modify the resistor array to juice the VIO regulator to 3.45V to see if that does the trick, but that's more of a curiosity.

Capacitors are very unlikely to be the barrier.

Vcore is ok, i guess. I`m looking in cpu-z under load. When it comes to capacitors, I would focus more on the northbridge voltage (3.3V?). The CPU power capacitors are sanyo and the northbridge power capacitors are "stone". Looking at the pictures of the board online, they are the first ones to leak while the sanyo still looks fine.

Nevertheless, now I will focus on RAM, users in this topic presented interesting facts about RAM on bx that I had no idea about. I need to check what I have

dulu wrote on 2024-08-03, 12:54:

I'm currently testing my PQI cube on my 440bx - rated at 133/CL3, running at 112/CL2 rock stable in prime and quake 2 loop

https://www.alldatasheet.com/datasheet-pdf/pd … EL/82443BX.html
112/CL2 - This is almost a standard combination for 440BX
The configuration registers have only 1 bit for the memory frequency
0- 66 MHz
1- 100 MHz
133 CL3 for 440BX is something like CL_2,509 for other chipsets (conditionally, the PC SDRAM does not have fractional CL, just 2 or 3)
When rounded, it seems closer to 3, but in fact - a very strict memory operating mode on the verge of the standard.

First, check the operation without RAM at all at 100 and 133 MHz.
If it beeps, then the motherboard itself starts at 133 frequency.
P.S. But not with Radeons, rarely any of them initialize with AGP 89 MHz.

dulu wrote on 2024-08-03, 13:00:

Looking at the pictures of the board online, they are the first ones to leak while the sanyo still looks fine.

It's easier to replace than to guess

I would focus more on the northbridge voltage (3.3V?)

Yes, together with memory and AGP.
Moreover, depending on the manufacturer, there may be a separate regulator for AGP and memory.
Or for everything, including PCI, like in Asus CUBX

PS.
You also need to update the BIOS to the latest version, even if it's beta.
Or, if someone modified

Some 440BX board don't set the 2/3 divider of FSB for the AGP (1/2 is not allowed by the north bridge itself, therefore "89 MHz AGP problem"), some don't set 33MHz for the south bridge.
The south bridge, unlike the north bridge, can operate at 133 bus.
This may also be a bug in this particular BIOS version.

Dont guess, measure. Put scope on Vcore, on VTT , on 3.3V.
Check if clockgen chip even generates 133MHz in the first place. I remember someone posting about Asus or Abit board with clockgen that should in theory do 133MHz but fitted one just couldt do it.

here is progress. I checked all the ram sticks I have - none of them are rated at 133/C2. I randomly started checking all the sticks one by one and the mobo started. Then, out of curiosity, I reconnected the PQI ram that I had been using all the time - the board also started.
Something must have changed because I tried to start the computer about 50 times. Before, I took the mobo out of the case to replace the retention bracket. Perhaps the reason was the atx20 plug not connected. This is the only plugin I haven't touched.
The computer starts up but I couldn't load Windows. Later I will keep trying to replace the GPU and when I get the 133/c2 stick, I will let you know if anything has changed. 😀

I got infineon 133/C2 ram, same as on @shevalier picture above. Windows freezes when loading. So it is weak bx piece or bad capacitors. i tried on pci vga.

dionb wrote on 2024-08-02, 21:48:

Whether you can overclock a board 33% beyond spec is a bit of a lottery, even with a high-quality board with late revision of a chipset famous for OC potential. I have a pile of MSI MS-6168 rev2 boards (late BX with onboard Voodoo3) and two of them reliably do 133MHz and two others utterly fail over about 112MHz.

As for the caps -if they are original, 24 year-old caps from the plague era are pretty suspect. You can check how suspect by measuring CPU core voltage under load. Is the board delivering the required voltage to the CPU? If so, I'd not primarily suspect the caps. If not, change the buggers.

Neither 6163 V1 can do it (or at least not mine) and I doubt my V2 board can do 133 FSB either. My 6168 won't do 133 either.

Oh, and the 6163 v1 I have is polymodded with 1200uF caps no less. It'll post at 133 but that's it, Windows is a no go.

I want this mobo to work with P3 1000/100 slot for fsb112, so the lack of fsb133 was not a problem for me. But after seeing the 2003 CPU test on tom`s hardware, I started to have doubts which combo would be faster, but there is no P3 1000/100 test.
- P3 1000/133
- P3 1000/100 @1120/112.
There are advantages and disadvantages. On the one hand, finding 1000/100 is much more difficult than 1000/133. Second one can be easily found for +-75$. (I'm not going to use slotket)
On the other hand, finding SDRAM 133/C2 on blue pcb is UBER hard.

Out of curiosity, I made theoretical calculations. Result - 1000/133 will be 4% faster in Q3. In practice it would probably look different, but it always sheds some light on the question

Which type of modded slotket do you use ? And does that combination work at fsb 133 on another board ?

modded slotket for mendocino celerons, 2 pins connected, one removed

Can you post a photo ?

here we go