Yes the solder blob can be easily sucked with a litz wire and I'll be back in original state within a minute when needed 😀

UPDATE: there are 2 "T" structures, each on one side. Both contains 0-ohm resistor...

Sphere478 wrote on 2024-04-09, 17:11:

Never bypass pullups/pulldowns like that if what was done is what I think was done. There is an example of someone doing this in the tillamook thread, I had to fix the procedure for them 🤣. Anyway, following their method it sometimes blows up a resistor.

Ah, I'd wondered if there might be a problem using boards that actually used the REGE pin, relieved it's just blowing a resistor. Must be quite a low value resistor for it to blow?

Sphere478 wrote:

I would have to look at this circuit closer to see what exactly is going on, but snufkin, could a smd resistor replace that solder blob?

Maybe, but it'll start to matter what value is used. A quick recap: On my Micron stick then there's already a 10k pull up on the input to the inverter, so that if nothing drives REGE (pin 147) then the output from the inverter will be low, so the stick will be registered. I removed the inverter and used two pads provided nearby to bypass the inverter. That means the 10k pull up is connected to the LE pin on the register chip, putting it in transparent mode. If a motherboard tries to drive the pin high then nothing happens. If a motherboard tries to drive the pin low then a current of around .33mA will flow. So nothing should break, but does mean the stick operates back-to-front: if the board tries to set the stick to registered then the stick will be unregistered. So there's a good chance that the stick won't work in a board that actually supports switching from registered to unregistered.

So, if the inverter isn't removed then the problem becomes how to neatly have the input to the inverter be default low, but without becoming a risk to the motherboard. That can either be done by lifting one end of the existing 10k pull up and connecting it to a convenient GND (I think that's what I'd do now). Or add a pull down resistor to the input that's strong enough to overcome the 10k pull up so that the input is still read as low.

The datasheet for the register chip on the Micron stick says that for a 2.5V input then an input must be under 0.7V. So I'll aim for 0.5V on a 3.3V supply. That'd mean a 2.8V (3.3V - 0.5V) drop over the existing 10k resistor, which would be .28mA. So if a pull-down was 1.79k then the input voltage would be 0.5V ( 0.5V / .28mA = 1.786kohm ). Using a smaller resistor will pull the input closer to GND, but also increase the current to a few mA (~3mA) if REGE is ever driven high.

Short version... Adding a 1k resistor between the input pin and GND should make the stick default to unregistered and keep the stick working as expected if the motherboard tries to drive REGE either high or low. Still not guaranteed to work in a motherboard that supports registered because a motherboard might rely on the stick default being registered, so could just leave REGE floating and expect that to set the stick to registered.

RayeR wrote on 2024-04-09, 20:37:

UPDATE: there are 2 "T" structures, each on one side. Both contains 0-ohm resistor...

That sounds right. Rotating both 0-ohm resistors seems to pick 3 different settings, I think for clock phase, one of which will give you the best timings. Clock signal comes in on one link, through a via, I think to a loop on an internal layer, through a via to the other side of the stick, through the second link, and back to the clock feedback. I'm assuming each of the 3 positions has a different internal loop length.

snufkin wrote on 2024-04-10, 17:01:

Ah, I'd wondered if there might be a problem using boards that actually used the REGE pin, relieved it's just blowing a resistor […]
Show full quote

Sphere478 wrote on 2024-04-09, 17:11:

Never bypass pullups/pulldowns like that if what was done is what I think was done. There is an example of someone doing this in the tillamook thread, I had to fix the procedure for them 🤣. Anyway, following their method it sometimes blows up a resistor.

Ah, I'd wondered if there might be a problem using boards that actually used the REGE pin, relieved it's just blowing a resistor. Must be quite a low value resistor for it to blow?

Sphere478 wrote:

I would have to look at this circuit closer to see what exactly is going on, but snufkin, could a smd resistor replace that solder blob?

Maybe, but it'll start to matter what value is used. A quick recap: On my Micron stick then there's already a 10k pull up on the input to the inverter, so that if nothing drives REGE (pin 147) then the output from the inverter will be low, so the stick will be registered. I removed the inverter and used two pads provided nearby to bypass the inverter. That means the 10k pull up is connected to the LE pin on the register chip, putting it in transparent mode. If a motherboard tries to drive the pin high then nothing happens. If a motherboard tries to drive the pin low then a current of around .33mA will flow. So nothing should break, but does mean the stick operates back-to-front: if the board tries to set the stick to registered then the stick will be unregistered. So there's a good chance that the stick won't work in a board that actually supports switching from registered to unregistered.

So, if the inverter isn't removed then the problem becomes how to neatly have the input to the inverter be default low, but without becoming a risk to the motherboard. That can either be done by lifting one end of the existing 10k pull up and connecting it to a convenient GND (I think that's what I'd do now). Or add a pull down resistor to the input that's strong enough to overcome the 10k pull up so that the input is still read as low.

The datasheet for the register chip on the Micron stick says that for a 2.5V input then an input must be under 0.7V. So I'll aim for 0.5V on a 3.3V supply. That'd mean a 2.8V (3.3V - 0.5V) drop over the existing 10k resistor, which would be .28mA. So if a pull-down was 1.79k then the input voltage would be 0.5V ( 0.5V / .28mA = 1.786kohm ). Using a smaller resistor will pull the input closer to GND, but also increase the current to a few mA (~3mA) if REGE is ever driven high.

Short version... Adding a 1k resistor between the input pin and GND should make the stick default to unregistered and keep the stick working as expected if the motherboard tries to drive REGE either high or low. Still not guaranteed to work in a motherboard that supports registered because a motherboard might rely on the stick default being registered, so could just leave REGE floating and expect that to set the stick to registered.

RayeR wrote on 2024-04-09, 20:37:

UPDATE: there are 2 "T" structures, each on one side. Both contains 0-ohm resistor...

That sounds right. Rotating both 0-ohm resistors seems to pick 3 different settings, I think for clock phase, one of which will give you the best timings. Clock signal comes in on one link, through a via, I think to a loop on an internal layer, through a via to the other side of the stick, through the second link, and back to the clock feedback. I'm assuming each of the 3 positions has a different internal loop length.

In that case I was lucky that that’s all that happened. It’s a bad thing to 0 ohm that bridge. You need a value high enough that it can survive and make a voltage divider. Which is how those work usually basically. (On BF pins on cpus)

snufkin wrote on 2024-04-10, 17:01:

That sounds right. Rotating both 0-ohm resistors seems to pick 3 different settings, I think for clock phase, one of which will give you the best timings. Clock signal comes in on one link, through a via, I think to a loop on an internal layer, through a via to the other side of the stick, through the second link, and back to the clock feedback. I'm assuming each of the 3 positions has a different internal loop length.

Is it needed to run (stable) the modules at nominal 133MHz speed?
Currently I don't have usage for the modules at this speed as my 440BX MBs don't take 512MB modules anyway (I use 3x 256MB double-sided modules there).
And in GA-5AA they run probably at lower speed, I don't know exactly what FSB2SDRAM ratio that Ali chipset use but I guess no more than 100MHz. I tested both modules and they passed test in GM (it took ~5 hours total). Also this Ali chipset is a piece of f* crap, as I mentioned in other thread, when FSB is set higher than 66MHz (I tried 75, 83, and 95MHz) then memory transfer rate (and also overal system performance) significantly degrade compared to 66MHz FSB. There's no explicit option to set some FSB2SDRAM ratio in SETUP (EDIT: according to M1541/M1542 datasheet it seems SDRAM runs synchronously with FSB, max official supported speed is 100MHz, Gigabyte added some overclocking). So currently I don't need to do any speed tuning on the modules. Do you tried all options? What's the best?

RayeR wrote on 2024-04-10, 20:21:

Is it needed to run (stable) the modules at nominal 133MHz speed? Currently I don't have usage for the modules at this speed as […]
Show full quote

snufkin wrote on 2024-04-10, 17:01:

That sounds right. Rotating both 0-ohm resistors seems to pick 3 different settings, I think for clock phase, one of which will give you the best timings. Clock signal comes in on one link, through a via, I think to a loop on an internal layer, through a via to the other side of the stick, through the second link, and back to the clock feedback. I'm assuming each of the 3 positions has a different internal loop length.

Is it needed to run (stable) the modules at nominal 133MHz speed?
Currently I don't have usage for the modules at this speed as my 440BX MBs don't take 512MB modules anyway (I use 3x 256MB double-sided modules there).
And in GA-5AA they run probably at lower speed, I don't know exactly what FSB2SDRAM ratio that Ali chipset use but I guess no more than 100MHz. I tested both modules and they passed test in GM (it took ~5 hours total). Also this Ali chipset is a piece of f* crap, as I mentioned in other thread, when FSB is set higher than 66MHz (I tried 75, 83, and 95MHz) then memory transfer rate (and also overal system performance) significantly degrade compared to 66MHz FSB. There's no explicit option to set some FSB2SDRAM ratio in SETUP (EDIT: according to M1541/M1542 datasheet it seems SDRAM runs synchronously with FSB, max official supported speed is 100MHz, Gigabyte added some overclocking). So currently I don't need to do any speed tuning on the modules. Do you tried all options? What's the best?

I think I had problems getting 133 / CL3 stable in the initial position and it was then stable (24 hours memtest) after repositioning those links. I suspect that the best position will depend on too many small things (buffer propagation delay, bus capacitance, ...) for there to be one good position, otherwise why have option? With the buffer still in place then the added delay means the stick is never going to be as fast as an unbuffered stick. But if you're already at the limit of the motherboard then none of that really matters.

Just adding some photos...

Ahhh get rid of that blob! Use a resistor 🤣

Looks like the PCB is the same as mine, so there's the option of moving the current 10k pull-up resistor that's on the back, by soldering on to the vias. Either above pins 147&148, or just to the side of the original pads. Easy enough to undo, leaves the inverter in place, should look reasonably neat, doesn't mean finding a single tiny resistor, and doesn't leave the small chance that the shorted-to-ground version ever finds its way on to another board.

Assuming I can still find the inverters I took off of mine then I think that's what I'd be doing.

[edit: Don't use the vias by the pins. I was being dumb and didn't realise that the via above pin 147/REGE connects to the pins on the front of the stick. So just use the vias to the side of the original pads]

I'll add some 0201 resistors on future order... 😀

i got my hands on different reg-sdram 512 and even 1g modules.

in short - every converted module runs fine on my Epox 3VSA2 which does not support ECC SDRAM.
all non-converted modules were stuck on C1 post code.

pics below:

pic 01. Tested modules

Pic 02: 1Gb SDRAM overview.

Pic 03: Invertor

Pic 04: connection

Pic 05: Success

Sphere478 wrote on 2024-04-12, 14:52:

Ahhh get rid of that blob! Use a resistor 🤣

Sure, I removed it sooner than I would expect. I tested one dual S370 sever board from HP and it failed to start with POST code 38h (Phoenix BIOS) with randomly picked SDRAM module i tried. Then I found a manual and read that MB requires the buffered ECC PC133 SDRAM only. So I remembered about my mod, put a dop of flux and gently touched blob with a solder tip and its't gone, mod reverted back. Then MB boots 😀 BTW it performs very heavy RAM testing at boot that may take severeal minutes with 1GB and even if pressed a key to skip it's still doing memory erasing that takes "only" 30s and cannot be disabled. Then extra wait for SCSI BIOS scannings LUNs, also couldn't be disabled. Rebooting such server many times was really a pain 😀

snufkin wrote on 2021-03-18, 20:23:

LE would be connected to pin 1467 (which should be NC)

This pin should be not connected (reserved) on very old motherboards or from manufacturers who did not read the datasheet.
First check the motherboard
- the chipset does not support FBDIMM, but pin 147 is a flow. Connect to ground on the motherboard
- the chipset supports FBDIMM, but pin 147 is a flow or high level (connected to 3.3V). So the problem is in the BIOS and you should moded the memory stick
- the chipset does not support FBDIMM, but pin 147 is grounded. The problem is in the BIOS, nothing will help.

I have two motherboards on I815
- Asus TUSL2-C
- Aopen MX3S
Both have 147 pin SDRAM grounded by design.
TUSL - works with FBDIMM in Transferred mode.
Aopen - only at first turn-on. Loads Windows, passes any tests.
Immediately after rebooting, it starts beeping about the lack of RAM.

Most likely, there is some error in Award BIOS that was not fixed by manufacturers who did not develop solutions for the server/workstation market.

PS. Knowing how Intel usually does things, I think that Intel D815EEA will happily start with FBDIMM and write "unsupported memory type, system halted. Contact technical support"

He-he, it worked
Old chipsets do not understand 256 MB strips with 8(9) memory chips, so they see half the capacity
The 0603 size resistor fits just perfectly.

So, you modded the motherboard to deactivate the registered chip? Am I understanding this correctly? Would this work on any board?

Oh crap. Bga 1gb SD sicks!

Where’d ya find those?

Sphere478 wrote on 2025-05-31, 04:50:

So, you modded the motherboard to deactivate the registered chip? Am I understanding this correctly? Would this work on any board?

on those motherboards where it is not connected to either ground or power
According to the latest standard, this contact should already be grounded on the motherboard = Transparency mode
But who "Reed The Fu...nny Manuals"

Sphere478 wrote on 2025-05-31, 04:50:

Oh crap. Bga 1gb SD sicks!

Where’d ya find those?

Where’d ya see those?
131072k= 128MB
It's Samsung vulgaris buffered ECC PC133
256 MB single rank stick (4+5 RAM chips on each side of the printed circuit board)
440BX and Via MVP3 see only half of the capacity.
The photo shows MVP3, which, according to the old standard, has this pin #147 "reserved", i.e. not connected anywhere.

shevalier wrote on 2025-05-31, 11:17:

on those motherboards where it is not connected to either ground or power According to the latest standard, this contact should […]
Show full quote

Sphere478 wrote on 2025-05-31, 04:50:

So, you modded the motherboard to deactivate the registered chip? Am I understanding this correctly? Would this work on any board?

on those motherboards where it is not connected to either ground or power
According to the latest standard, this contact should already be grounded on the motherboard = Transparency mode
But who "Reed The Fu...nny Manuals"

Sphere478 wrote on 2025-05-31, 04:50:

Oh crap. Bga 1gb SD sicks!

Where’d ya find those?

Where’d ya see those?
131072k= 128MB
It's Samsung vulgaris buffered ECC PC133
256 MB single rank stick (4+5 RAM chips on each side of the printed circuit board)
440BX and Via MVP3 see only half of the capacity.
The photo shows MVP3, which, according to the old standard, has this pin #147 "reserved", i.e. not connected anywhere.

Another user posted above.

I just scored one when I went looking for it! W00t! I’ll mod mine also probably 😀

Can add it to my BGA collection