snufkin wrote on 2022-01-04, 12:04:

Huh. Ok. Either I've miscounted and that's not pin 126. Or it's not routed on the motherboard. In which case how does it cou […]
Show full quote

Sphere478 wrote on 2022-01-04, 01:48:

No connection between slots on that pin on p5a 1.04

Huh. Ok. Either I've miscounted and that's not pin 126. Or it's not routed on the motherboard. In which case how does it count to 1GB when it's missing A12, so half the Row addresses are missing?

Maybe it doesn't know that it can't, and it's actually wrapping around. If it reads and believes the SPD then it knows that it needs to test 1GB (128M locations * 8 bytes at each location). If it just does a simple test (write value, read value back) then it won't be able to tell if the location wraps around when it gets past half way. It'd need to do a test where it first wrote non-repeating data to all the addresses before reading back. So it could be doing a simple test and passing, then it writes some actual data to RAM, overwriting something important, and crashes. Maybe the BIOS on the ga-5aa stops at 512MB not because it gets an error, but just because it's programmed to not count past 512MB because it can't address it.

Can you measure the resistance from that pin 126 to pin 127 (Ground) and to pin 124 (Vcc)? Maybe they just pull it high or low.

Still, if that is pin 126 and it's not routed, then that'd explain why 1GB won't work. Which is annoying. Getting the A12 line out from the 1541 would be tricky.

This makes sense. On my tyan s1564 it would count 768mb of simms and boot, but then randomly crash but memtest would go on for days without error.

It was wrapping back around as you say.

Pop in 512mb and it works just fine (max supported/addressable)

If they were pulled high or low then they would have connection to each other.

Sphere478 wrote on 2022-01-04, 12:57:

This makes sense. On my tyan s1564 it would count 768mb of simms and boot, but then randomly crash but memtest would go on for days without error.

At least some of the memtest tests should fail if it's wrapping around. I think. Been a while since I read what the tests actually are.

If they were pulled high or low then they would have connection to each other.

Might be through a big value resistor, so the connection from one to the other would be twice that resistance. What's the range on your meter?

I'm now wondering what's responsible for driving any unused address lines. Or if it's just a matter of not putting oversized DIMMs in.

snufkin wrote on 2022-01-04, 14:29:

At least some of the memtest tests should fail if it's wrapping around. I think. Been a while since I read what the tests actu […]
Show full quote

Sphere478 wrote on 2022-01-04, 12:57:

This makes sense. On my tyan s1564 it would count 768mb of simms and boot, but then randomly crash but memtest would go on for days without error.

At least some of the memtest tests should fail if it's wrapping around. I think. Been a while since I read what the tests actually are.

If they were pulled high or low then they would have connection to each other.

Might be through a big value resistor, so the connection from one to the other would be twice that resistance. What's the range on your meter?

I'm now wondering what's responsible for driving any unused address lines. Or if it's just a matter of not putting oversized DIMMs in.

I was getting connection only in megaohm range

Sphere478 wrote on 2022-01-04, 23:08:

I was getting connection only in megaohm range

Ok... so no pull up or down then. So almost certainly just NC. It's probably worth double checking that I've counted correctly. When you've got a moment, could you check that pin 127 (one further away from the middle) connects to ground? And that pin 124 (just the other side of the middle divider) connects to 3.3V? And that pin 125 (CK1) *isn't* connected to 3.3V (that last one should check that I'm not on completely the wrong side).

If it is just NC then it could just float around depending on how much electrical noise is around, so sometimes read high or low. That'd also cause strange errors.

But if this is right and MA[14] wasn't routed out from the 1541 then of course the hunt becomes whether there ever was a board that did route it out. I assume ALi had at least an engineering prototype.

Don't suppose you know anyone who can reball BGAs? Then you could remove the 1541, wire on to the MA[14] pad on the 1541 (use some relay coil wire), reball the rest and reflow it, then connect the wire to that pin 126 on the SDRAM. Impedance and length would be wrong, so almost guaranteed not to work...

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

[edit: probably useful to note up here that I eventually figured out that all I'm doing is using the REGE pin 147 for its intend […]
Show full quote

[edit: probably useful to note up here that I eventually figured out that all I'm doing is using the REGE pin 147 for its intended purpose, to disable the register function, but keep the signals buffered]

(Short version: I accidentally got some registered RAM and converted it to unregistered)
I've just registered as I thought this might be of interest, and in case anyone has any suggestions about testing whether ECC is functioning on an Abit KA7 motherboard. Apologies if I break any etiquette.

PC133r.jpg
Just been spending some time getting an Abit KA7 working again and was having trouble with the RAM. As part of an overly long story (including discovering that all the 3.3V on this board is regulated locally from PSU 12V, it doesn't use the PSU3.3V at all) I thought I'd try some ECC RAM (although I've since seen that the KX133 may not actually have implemented ECC correction anyway).

Problem was that I wasn't careful enough, and ended up getting some Micron PC133r probably from a Compaq server. And the KA7 doesn't support registered RAM (I think the extra clock tick delay mean data reads and writes don't line up). Having a quick look around it seemed that converting from registered to unregistered was a non-starter, so I'd have to look for some unregistered ECC ram.

That wasn't very common though, unlike all the ex-server stuff. So I had a closer look at the register chips on the sticks I had. In particular the 74ALVCF162835. Looked up a datasheet for it and in the second paragraph saw the following sentence:
"The device operates in Transparent Mode when LE is held HIGH."
PC133r_Register.jpg

Well that sounded interesting. If the problem with using registered RAM was the delay caused by the register, then a non-clocked mode might help. I assumed (wrongly) that they'd have just tied the pin (28) low on each of the 3 registers, so I'd have to lift each pin and tie it high. Turns out all the pins were connected together, and I eventually found they were all connected to a pin on a small 5 pin device. That turned out to be a single bit inverter, the input of which was pulled high via a 10k resistor. Which means the output would be low, and the registers in clocked mode. But why go to that trouble? Turns out the input was also connected to pin 1467 on the DIMM, which I think is supposed to be NC. So it looks as though Micron wanted the motherboard to be able to switch the RAM between registered or unregistered. Which sounded hopeful.

Just above the inverter were a couple of pads for an 0805 sized (I think) component and it turns out they go between the input and output of the inverter. So if I removed the inverter and shorted across the pads, then LE would be connected to pin 1467 (which should be NC) and a weak pull up to 3.3V. So I went ahead and did that.
PC133r_LE_Control.jpg

I guessed that I'd need to edit the SPD data stored on the DIMM, otherwise the motherboard would see that it said it was registered and not try to use the RAM. But I thought I'd just quickly try it first. And it booted. I assume BIOS writers do as little as possible to make things work, and checking to see whether or not you might be able to use the RAM, when you can just try it, is an extra step not needed.

So I now have buffered but unregistered ECC RAM. Of the 4 sticks I got, the 3 128MB ones all work at PC133/CL2 (tested one at a time, haven't tried installing all at the same time yet), but the 512MB one fails memtest86+ when run at PC133/CL3. It is stable at PC100/CL2 though.

Just tried installing edac_util in linux, but that just says "no memory controller data found". So probably no go with testing the ECC bit, unless anyone has any suggestions? I'll go with the '24 hours of memtest' instead.

So, for some registered RAM and some motherboards, converting from registered to unregistered just means removing one component and adding a short.

i have the same brand sticks but are 256mb sticks, and has 5memory chips on one side, and 4 memory chips on the other side
but it looks all different from your photo, if i can post some HQ image maybe you can help me what components should i bridge and remove?

BLockOUT wrote on 2022-02-26, 04:02:

i have the same brand sticks but are 256mb sticks, and has 5memory chips on one side, and 4 memory chips on the other side
but it looks all different from your photo, if i can post some HQ image maybe you can help me what components should i bridge and remove?

Would need higher resolution pictures, but it looks like the inverter for the REGE pin is on the front between the two right-hand DRAM chips. Looks like it has a couple of bare pads next to it. If so, it probably means removing the inverter and shorting those two pads. You should be able to check with a multimeter for connections between there and pin 147 and the register enable pins on the register chips. Easiest done if you can find datasheets for the register chips and the inverter.

It also looks like what I think is the rough clock phase control is on the back between the two register/buffer chips, which may need to be adjusted to get the best speed/stability out of it.

[edit: Actually, looks like the bypass short is on the back, next to what looks like a pull up resistor.]

snufkin wrote on 2022-02-26, 13:21:

BLockOUT wrote on 2022-02-26, 04:02:

i have the same brand sticks but are 256mb sticks, and has 5memory chips on one side, and 4 memory chips on the other side
but it looks all different from your photo, if i can post some HQ image maybe you can help me what components should i bridge and remove?

Would need higher resolution pictures, but it looks like the inverter for the REGE pin is on the front between the two right-hand DRAM chips. Looks like it has a couple of bare pads next to it. If so, it probably means removing the inverter and shorting those two pads. You should be able to check with a multimeter for connections between there and pin 147 and the register enable pins on the register chips. Easiest done if you can find datasheets for the register chips and the inverter.

forget that one, i got some 512mb power sticks
but sadly they are completely different from what you posted

here are some macro high resolution pictures from the stick, notice i had to cut it in pieces because the camera is really bad but you can zoom a lot
picture zip: https://www.mediafire.com/file/b1sq8wgq5arbad … hsides.zip/file

what i saw is that your 5 pin leg thing on yours on mine is on one side, the middle leg goes into a via that goes to the other side of the memory, and then goes into another component

i was trying to check if there was continuity between the middle leg and any blank unsoldered component pads but there was no continuity

any advise?

My guess is that the side of the 10k resistor (marked 103) you've marked with the arrow will connect to pin 147 of the edge connector. The other side of the resistor probably connects to Vcc. That would mean that if pin 147 isn't driven by the motherboard then the middle pin of the 5 pin chip (input A) will be pulled high. That chip is probably in inverter, so the output pin (bottom-right, Y) will then go low so that the bus drivers default to registered mode. I think the easiest way to convert it is to move the resistor so that it connects to Gnd instead of Vcc. That will pull the input to the inverter low, so the output will be high, so the bus drivers will be in transparent mode (registers turned off).

Check that the wiring matches what I've marked in the photo. If it's all ok, then I'd probably try lifting the 10k resistor, tape off the Vcc pad, then put the resistor back using what I think is the nearby Gnd via. Be careful not to short Vcc to Gnd. Assuming that the motherboard doesn't drive the pin (it probably doesn't) then that should work.

I don't see the 'T' structure anywhere, so there may not be the option of fiddling with what I think is the phase adjust to get the best timings.

(yes, you are right on all in your picture!,

top side of 103 has continuity to the REGE golden connector)
+ VCC points have continuity on the left picture
+ Ground points have continuity on the left picture
+ Y point on left picture has continuity with Y on the right picture
+ Grounds on the right pictures have continuity
+ VCC points on the right picute have continuity!

snufkin wrote on 2022-08-09, 08:33:

My guess is that the side of the 10k resistor (marked 103) you've marked with the arrow will connect to pin 147 of the edge conn […]
Show full quote

My guess is that the side of the 10k resistor (marked 103) you've marked with the arrow will connect to pin 147 of the edge connector. The other side of the resistor probably connects to Vcc. That would mean that if pin 147 isn't driven by the motherboard then the middle pin of the 5 pin chip (input A) will be pulled high. That chip is probably in inverter, so the output pin (bottom-right, Y) will then go low so that the bus drivers default to registered mode. I think the easiest way to convert it is to move the resistor so that it connects to Gnd instead of Vcc. That will pull the input to the inverter low, so the output will be high, so the bus drivers will be in transparent mode (registers turned off).

ConvertSDRAM.jpg
Check that the wiring matches what I've marked in the photo. If it's all ok, then I'd probably try lifting the 10k resistor, tape off the Vcc pad, then put the resistor back using what I think is the nearby Gnd via. Be careful not to short Vcc to Gnd. Assuming that the motherboard doesn't drive the pin (it probably doesn't) then that should work.

I don't see the 'T' structure anywhere, so there may not be the option of fiddling with what I think is the phase adjust to get the best timings.

Like this?
1) i remove the 5 leg thing
2) i remove the 103
3) i place 103 on the other side, on the 2 bottom left legs

snufkin wrote on 2022-08-09, 08:33:

Check that the wiring matches what I've marked in the photo. If it's all ok, then I'd probably try lifting the 10k resistor, tape off the Vcc pad, then put the resistor back using what I think is the nearby Gnd via. Be careful not to short Vcc to Gnd. Assuming that the motherboard doesn't drive the pin (it probably doesn't) then that should work.

I don't see the 'T' structure anywhere, so there may not be the option of fiddling with what I think is the phase adjust to get the best timings.

sorry missed this comment.

yea your idea is even better, i can lift the 103, then use some green mask with UV light over the vcc area, then place back the 103 and use a little wire to conect the other end to the near ground point.
i think that is more elegant, But what about the 5 leg inverter thing on the other side, should i leave it on the pcb or should i remove the inverter?

BLockOUT wrote on 2022-08-09, 15:21:

But what about the 5 leg inverter thing on the other side, should i leave it on the pcb or should i remove the inverter?

That stays. The 10k resistor is on the input pin of the inverter, normally pulling it high, unless the input is over-ridden by a signal coming in on pin 147 from the motherboard. By connecting the resistor to Gnd instead of Vcc then the output of the inverter will change to being normally high, unless over-ridden by the motherboard, which should put the bus drivers in unregistered mode. The unregistered SDRAM pinout has pin 147 as not connected, so chances are that the motherboard can't change the input to the inverter, so the SDRAM will be set to unregistered.

The other option is to remove the inverter and short the input and output pads. That's sort of what I did originally (on my stick there was space for a 0-ohm SMD link to short the input and output) until I finally found out about the REGE pin 147, at which point I figured it was neater to just change the pull-up/down resistor.

[edit: incidentally, there are a couple of empty pads to the left of the inverter. If you haven't already, might be worth checking if they go to the inverter input (middle-left) and output (bottom-right). If they do, then removing the inverter and shorting those pads might also be fairly neat. I still prefer changing the pull-up/down resistor as it's easier to undo]

snufkin wrote on 2022-08-09, 16:27:

BLockOUT wrote on 2022-08-09, 15:21:

But what about the 5 leg inverter thing on the other side, should i leave it on the pcb or should i remove the inverter?

[edit: incidentally, there are a couple of empty pads to the left of the inverter. If you haven't already, might be worth checking if they go to the inverter input (middle-left) and output (bottom-right). If they do, then removing the inverter and shorting those pads might also be fairly neat. I still prefer changing the pull-up/down resistor as it's easier to undo]

thanks, no those empty pads don´t have continuity to any of the 5 legs.
there are 2 pads on the other side but only one has continuity to ground.

probably this week i will work on desoldering the 103 but i want to be careful, i bought a desoldering tweezer but its so big that i cant even fit it between the memory and the 103. I don´t understand how they don´t sell these things super micro small for smd components. And i also need to find a cheap motherboard just in case because my pentium3 motherboards are high quality and difficult to find replacement.
then run memtest for some hours

BLockOUT wrote on 2022-08-10, 00:13:

I don´t understand how they don´t sell these things super micro small for smd components.

A few reasons... small bits of metal are thin and lack the cross sectional area to transfer significant amounts of heat, you'd need the top end of it glowing to maintain soldering temp at the board. Small bits of metal also lack much heat capacity, so even if you get it hot enough at the moment to theoretically melt solder, the moment you touch it, it's dumped most of the heat it was carrying and has frozen to it with the solder. Then also, small pieces of metal heated frequently are prone to getting eaten by the heat and oxidation, basically you'll use it twice and need new tips. So if they made a small one, you'd probably like it even less than the big one.

Thanks for this thread!
I just digged out 2 huge registered DIMM modules IBM MT18LSDT6472G-133B1 (looks similar to those on 1st post) that was lying there spare for years without usage. So I did a simple mod by adding a solder blob just on the SMD invertor chip between input pin (REGE) and neighbour GND and yeah it booted in my test GA-5AA rev. 2.2 😀 Just running Goldmem there to see if they are OK (Memtest 4.x always hangs on this MB for some reason so I use GM that I found much more stable on older MBs). Crazy that I have now more memory on S7 than I ever had on S370/440BX boards (it should support 1GB but all MBs I have has only 3 slots).
BTW I set error checking in SETUP to ECC - does ALi chipset really supports ECC? How can I read some ECC event counts?

Glad it worked. Probably best to avoid using those sticks in a motherboard that supports registered RAM, just in case the motherboard tries to drive the pin high when it's shorted to ground. Dunno about testing ECC, from what I read the VIA chipset on my KA7 claimed to support ECC but others found it doesn't. But I don't know how that was tested. Could maybe try removing the terminating resistor pack for a couple of the check bit pins (21&22 looks easyish to get to), effectively disconnecting those ECC bits, then seeing if you get any memory errors?

I currently don't have any server MB that use registered SDRAM so I don't care much but I agree that it would be cleaner to solder some 0402 or 0201 tiny resistor instead of solder blob, maybe later, I just marked a note on a module sticker...
BTW there's also PLL circuit in this modules and I seen some "T" structure, not sure how to play with it and what effect it has. On current MB the modules runs on lower than max. speed anyway...
I read that in general some MBs may accept ECC modules but don't really use the ECC feature, there must be combined support of ECC RAM, chipset and MB design/BIOS? to make it all working. I think that Linux may report ECC evenets, maybe in dmesg or somewhere in /proc/...?
It should be enough to push some timings to the limit to trigger some ECC events (or lower RAM voltage if MB supports it)...

I think the T structure acts as a sort of phase control for the clock. On mine there was a matching structure on the front and back, both with a 0 ohm link between the centre pad and one of the three ends. As long as the link was in the same position on both sides then the memory worked, but one position allowed for faster timings than the other two. There's only 3 possible positions so it didn't take too long to test.

In my case it looked rather like a capacitor? not sure now, later I take some photo and measure what the component is...

snufkin wrote on 2024-04-09, 14:38:

Glad it worked. Probably best to avoid using those sticks in a motherboard that supports registered RAM, just in case the motherboard tries to drive the pin high when it's shorted to ground. Dunno about testing ECC, from what I read the VIA chipset on my KA7 claimed to support ECC but others found it doesn't. But I don't know how that was tested. Could maybe try removing the terminating resistor pack for a couple of the check bit pins (21&22 looks easyish to get to), effectively disconnecting those ECC bits, then seeing if you get any memory errors?

Yes, this is a huge mistake.

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.

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?

I’m sure RayeR was trying to make it reversable was the idea to not loose the inverter in a drawer somethere.