VOGONS


Converted registered SDRAM

Topic actions

Reply 21 of 73, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Re: ga-500 super socket 7 motherboard

Okay, here is my specific build thread, I just ordered three sticks of the ram I posted.

Will try and mod it.

Are you able to show on the pic what has to be done? Or do you need a higher res pic?

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 23 of 73, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie
Sphere478 wrote on 2021-11-08, 20:42:

Are you able to show on the pic what has to be done? Or do you need a higher res pic?

Jumping straight in with all 3? Exciting. Have you got a multimeter to hand? It's going to need some measuring to try and figure out what's connected to where, particularly that pin 28 on the register chips. It looks like these are multilayer PCBs, with signal on the inside layers, so photos might help a bit but probably won't give enough information. Pictures of the back would probably be useful as there'll most likely be a couple of register chips there.

luckybob wrote on 2021-11-08, 19:16:

this looks and feels like heresy.

But really it's just I can't believe it's not heresy?

luckybob wrote on 2021-11-08, 20:51:

Ok, this is officially heresy now.

Ooh, is there a badge?

Reply 24 of 73, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
luckybob wrote on 2021-11-08, 20:51:

Ok, this is officially heresy now.

3gb of ram on a k6 3+ is my kinda heresy! :p

snufkin wrote on 2021-11-08, 22:48:
Jumping straight in with all 3? Exciting. Have you got a multimeter to hand? It's going to need some measuring to try and fig […]
Show full quote
Sphere478 wrote on 2021-11-08, 20:42:

Are you able to show on the pic what has to be done? Or do you need a higher res pic?

Jumping straight in with all 3? Exciting. Have you got a multimeter to hand? It's going to need some measuring to try and figure out what's connected to where, particularly that pin 28 on the register chips. It looks like these are multilayer PCBs, with signal on the inside layers, so photos might help a bit but probably won't give enough information. Pictures of the back would probably be useful as there'll most likely be a couple of register chips there.

luckybob wrote on 2021-11-08, 19:16:

this looks and feels like heresy.

But really it's just I can't believe it's not heresy?

luckybob wrote on 2021-11-08, 20:51:

Ok, this is officially heresy now.

Ooh, is there a badge?

I have a fluke and a soldering iron and hot air gun but I’m really not experienced or have the equipment for small soldering like this though, the smallest I’ve done is pin mods on a tillamook

But I will try. Probably on one at first of course.

I can post readings from the meter and better pics when I get them

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 25 of 73, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Okay, here they are!!! what do you think?

These are some cool looking sticks!!! haha,

hope this works! Because they don’t work as they are 🙁

Attachments

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 26 of 73, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

Right, fingers crossed, let's check for the easy option first. Measurements to take:
1) Measure the resistance between pin 28 (!LE) on the 3 16834 register chips. With a bit of luck they'll all be connected together and should measure as under a couple of ohms.
2) If they are connected together then on one of the chips measure between pin 28 (!LE) & pin 29 (GND), and...
3) pin 28 & pin 35 (Vcc). Hopefully one of those (probably !LE to Vcc) will be connected through a resistor (on the sticks I modded it was a 10k resistor). If the pin 28s aren't connected together then do this on each 16834 chip.
4) Measure from pin 28 to pin 147 on the edge connector. It may be nothing, but (similar to the pin 1467 on my stick) pin 147 should be not connected, and yet there's a trace coming from it. There's a chance that Google used pin 147 as the latch enable/disable signal.

What I'm hoping is that the 3 !LE pins are connected together and pulled high to Vcc through a ~10k resistor. If that's true then it should mean you can make the simple mod of adding a small resistor.

[edit: actually many edits... turns out I can't count and the latch enable input on my sticks isn't pin 146 after all. It's pin 147, the same as the Google stick, which I think increases the likelihood that it should be possible to make a simple mod]
[and now that I've got the right pin number I found this: http://www.hardwarebook.info/168_pin_DIMM_SDRAM which says that pin 147 on registered sticks is REGE, which sounds like register enable to me, so it seems like this is actually all working to spec. I assume server BIOSes could be set to turn off the register if the system needed lower latency]
[and REGE led me to https://media-www.micron.com/-/media/client/g … bf6b840c0fa8b99 which on page 6 is explicit that the stick can be used in either registered or buffered modes. Things I never knew].
[I wonder why motherboards that didn't support registered RAM didn't just connect pin 147 to GND. Unbuffered sticks wouldn't care as it'd be NC on the stick, Registered sticks would then work as just buffered and have a chance of working. Probably reduced customer support issues by just having registered RAM not work rather than maybe work].

Attachments

Reply 27 of 73, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Should I just try soldering a wire between 147 and ground? Do you have a stick you could try it on, see if it works? Maybe a small jumper or switch could be installed also.

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 29 of 73, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie
Sphere478 wrote on 2021-11-11, 18:26:

Should I just try soldering a wire between 147 and ground? Do you have a stick you could try it on, see if it works? Maybe a small jumper or switch could be installed also.

I think it'd be better to figure out what's connected to what on the sticks you have before starting to add any jumpers. We don't know if the ones I have are wired up the same, and it might show an easy place to make a mod. I don't think you'll be able to solder directly to pin 147 and still get the stick in the slot, but if there's a pull up resistor on the trace then that might be a good place to add a stronger pull down resistor. I'm hoping not to have to make any permanent alterations, such as cutting traces.

maxtherabbit wrote on 2021-11-11, 18:30:

EXTERMINATUS

Any chance we can see if this works before all the purging begins?

Reply 30 of 73, by luckybob

User metadata
Rank l33t
Rank
l33t

The only end result of this thread is a pile of dead parts.

I feel like I'm watching a car wreck in slow motion. Except the cars are retro parts.

It is a mistake to think you can solve any major problems just with potatoes.

Reply 31 of 73, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

Entropy gets everything eventually. That's a proper slow motion crash.

In the meantime I think I've figured out the easy place to make the mod. Turns out there's something that looks very like where a 5 pin inverter could go, which would let the manufacturer use registers with the !LE function inverted (like the ones on the Micron registered sticks). If I'm right then it should be a simple case of attaching a resistor between where the output would be and the GND pin.

32Chip_SDRAM_Back_InverterPads.jpg
Filename
32Chip_SDRAM_Back_InverterPads.jpg
File size
35.95 KiB
Views
702 views
File license
Fair use/fair dealing exception

So, check the resistance between Y (the output) and Vcc. Hopefully it's around 10k. Also check the resistance from Y to pin 28 on the register, that should be around 0 ohms. If the resistance from Y to Vcc is around 10k then put a 1k resistor between Y and GND. That should then pull pin 28 low enough to disable the register, without risking damaging the motherboard if for some odd reason it's trying to drive pin 147 high.

Reply 32 of 73, by luckybob

User metadata
Rank l33t
Rank
l33t

I mean, the memory controllers on the north bridge are NOT meant to handle the load of 1gb sticks of ram. That is the *ENTIRE* reason registered ram exists.

You guys are putting 1 ton of bricks in a truck that can only handle 1/4 ton.

Quite literally.

This will only end in disaster.

It is a mistake to think you can solve any major problems just with potatoes.

Reply 33 of 73, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie
luckybob wrote on 2021-11-11, 21:21:

I mean, the memory controllers on the north bridge are NOT meant to handle the load of 1gb sticks of ram. That is the *ENTIRE* reason registered ram exists.

AIUI, all this does is turn the register in to a buffer, it doesn't expose the address/control of the DRAM chips directly to the memory controller. The datasheet for the register shows a final buffer stage regardless of the LE control, and I haven't seen anything in the datasheet that says the input characteristics change when it's in buffer mode.

[edit: Oh, and the clock goes through a PLL, so that's effectively buffered as well. Maybe there'll be a problem with the draw on the 3.3V supply, might be worth keeping an eye on by measuring the voltage across one of the capacitors on the RAM and seeing if it drops when going from 1 stick to 2.]

So the load (at least for the up to 54 buffered lines) should be less than it would be for an unbuffered stick, the overall timings won't be quite as good as with unbuffered (there's a propagation delay in the buffer), but the latency will be better than registered. At the moment I don't see where the electrical risk is. I can easily imagine problems with BIOS and cacheing limits stopping this from working, but (as long as the REGE input is pulled low through a resistor to limit any current on that line) I don't think anything should actually break.

Last edited by snufkin on 2021-11-11, 22:31. Edited 1 time in total.

Reply 34 of 73, by weedeewee

User metadata
Rank l33t
Rank
l33t
luckybob wrote on 2021-11-11, 21:21:
I mean, the memory controllers on the north bridge are NOT meant to handle the load of 1gb sticks of ram. That is the *ENTIRE* r […]
Show full quote

I mean, the memory controllers on the north bridge are NOT meant to handle the load of 1gb sticks of ram. That is the *ENTIRE* reason registered ram exists.

You guys are putting 1 ton of bricks in a truck that can only handle 1/4 ton.

Quite literally.

This will only end in disaster.

You made your warning, now please stop it with these negative waves.

Right to repair is fundamental. You own it, you're allowed to fix it.
How To Ask Questions The Smart Way
Do not ask Why !
https://www.vogonswiki.com/index.php/Serial_port

Reply 35 of 73, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
snufkin wrote on 2021-11-11, 18:55:
I think it'd be better to figure out what's connected to what on the sticks you have before starting to add any jumpers. We don […]
Show full quote
Sphere478 wrote on 2021-11-11, 18:26:

Should I just try soldering a wire between 147 and ground? Do you have a stick you could try it on, see if it works? Maybe a small jumper or switch could be installed also.

I think it'd be better to figure out what's connected to what on the sticks you have before starting to add any jumpers. We don't know if the ones I have are wired up the same, and it might show an easy place to make a mod. I don't think you'll be able to solder directly to pin 147 and still get the stick in the slot, but if there's a pull up resistor on the trace then that might be a good place to add a stronger pull down resistor. I'm hoping not to have to make any permanent alterations, such as cutting traces.

maxtherabbit wrote on 2021-11-11, 18:30:

EXTERMINATUS

Any chance we can see if this works before all the purging begins?

Just conversing while not home, thinking, pondering I will test.

Maybe modding the motherboard would also make sense

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 36 of 73, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

The 28 (!LE) pins are all under a ohm to one another

28k !LE to gnd

9.89k !le to vcc

snufkin wrote on 2021-11-11, 20:36:
Entropy gets everything eventually. That's a proper slow motion crash. […]
Show full quote

Entropy gets everything eventually. That's a proper slow motion crash.

In the meantime I think I've figured out the easy place to make the mod. Turns out there's something that looks very like where a 5 pin inverter could go, which would let the manufacturer use registers with the !LE function inverted (like the ones on the Micron registered sticks). If I'm right then it should be a simple case of attaching a resistor between where the output would be and the GND pin.
32Chip_SDRAM_Back_InverterPads.jpg

So, check the resistance between Y (the output) and Vcc. Hopefully it's around 10k. Also check the resistance from Y to pin 28 on the register, that should be around 0 ohms. If the resistance from Y to Vcc is around 10k then put a 1k resistor between Y and GND. That should then pull pin 28 low enough to disable the register, without risking damaging the motherboard if for some odd reason it's trying to drive pin 147 high.

I don’t have any smd resistors laying around. Or really the tools for working with such small things. But maybe this is a job for a conductive pen?

It sounds like since this stick doesn’t have a inverter that we have to mod the sticks yeah?

9.89k between y and vcc

1 ohm y to 28 !LE

Seems like I should find that 10k resistor? Or have we already? And remove it, then put a 10k to gnd or use a conductive pen?

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 37 of 73, by snufkin

User metadata
Rank Oldbie
Rank
Oldbie

No reason to mod the motherboard, so let's leave that untouched. Could you check the resistance from REGE to Y? I'm expecting it to be around 0 ohms. If you can, remove that 0 ohm link shown in the last picture and recheck the resistance from REGE to Y. After removing the link It should be disconnected which means there's no worry about the motherboard any more. At that point any means of shorting Y to GND should be safe, so either try using the 0 ohm link you will have just removed, or if you have a conductive pen you can use that.

On removing SMD components, don't apply force to lift them, you might damage traces and vias. Depending on the size of the soldering iron tip then you may be able to melt both ends at the same time, then use a pair of small tweezers to just lift it up.

The other option would be to find the 10k pull up resistor, remove it and put it between Y and GND. Not sure where it is. You'll be looking for a single resistor marked 103. I'm guessing it's one of the ones on the top, so measure the resistance between pin 28 on one of the registers and the various resistor pads. Eventually you should find one that reads close to 0 ohms. The other side of the resistor should be connected to Vcc. If you can find that, then you've found the pull up resistor.

Reply 38 of 73, by zapbuzz

User metadata
Rank Oldbie
Rank
Oldbie
luckybob wrote on 2021-11-11, 18:58:

The only end result of this thread is a pile of dead parts.

I feel like I'm watching a car wreck in slow motion. Except the cars are retro parts.

Well they haven't found a way to set the t-model fords on fire yet then
maybe if they add nitro 🤣

Reply 39 of 73, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

Rege to y is 0.3 ohm

I’m not seeing such a resistor, you said at the top? Hard to read them though. Which ones do you suspect?

I can probably handle removal of a resistor, I have tweezers and a hot air gun. But past that, I don’t feel I have the right tools for solder work this small.

Which conductive pen should I use?
I only know of them from the athalon days but never bought or ordered one myself.

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)