When a DIMM RAM stick gets cost-optimized just that little bit too far?

[SilverSolder]

These are 128MB DIMM RAM sticks for HP printers...   the aftermarket version on top, the HP original on the bottom.

The aftermarket version works...  as long as you only put one aftermarket RAM stick in the printer at a time (it has 2 free slots)!  If you put aftermarket DIMMs in both slots, the printer stops on boot with a DIMM error.

The original equipment DIMMs work fine with both slots populated...


Looking closer at them, one difference seems to be more liberal use of capacitors on the OEM memory.  Could the paring down of the capacitors on the aftermarket stick explain why they are marginal?

[amyk]

I doubt this has anything to do with signal integrity. DIMMs are standardised --- even the PCB layouts --- and you can find all the details if you look around on jedec.org. This is the appropriate one: https://www.jedec.org/standards-documents/docs/module-40408

128MB of PC100 doesn't tell you what the organisation of the bits is. The spec above says that modules may have 1 or 2 banks, so if your printer can only use two banks in total, you can't use two 2-bank DIMMs.

Some PC motherboards have similar limitations.

[JohnnyMalaria]

The aftermarket has more caps and no resistors. But these images are not the same side of the stick. The HP stick doesn't have resistors on the other side.

[SilverSolder]

I don't have a picture of the other side of the 3:rd party module, but it looked close to the same as the side shown.  The HP module looks nearly the same on both sides as well.

The aftermarket has more caps and no resistors.

I count what looks like 17 caps on one the HP module, plus what looks like a large tantalum cap.  I see 8 caps on the aftermarket one.  Am I missing something obvious?



I guess another way to look at it is -  Why would the designer of the HP module specify so many caps, if they are not necessary for proper functioning?

[SilverSolder]

I doubt this has anything to do with signal integrity. DIMMs are standardised --- even the PCB layouts --- and you can find all the details if you look around on jedec.org. This is the appropriate one: https://www.jedec.org/standards-documents/docs/module-40408

128MB of PC100 doesn't tell you what the organisation of the bits is. The spec above says that modules may have 1 or 2 banks, so if your printer can only use two banks in total, you can't use two 2-bank DIMMs.

Some PC motherboards have similar limitations.

The HP manual says the printer supports 2x 128MB DIMMS, and indeed the printer works perfectly with 2x HP DIMMs...

Somehow, the aftermarket DIMMs are a clear Fail...   I was just wondering if there is such a thing as not having enough bypass capacitors and this could cause issues?

[helius]

When you are dealing with memory subsytems, there's no such thing as "just a 128MB DIMM". The stick has an internal organization built out of the chips that are on it. Some sticks are "double ranked" or "2R" and are recognized by the memory controller as two separate memory devices. That is most likely what is going on with your DIMM.

Another possibility is that the refresh depth is different and the memory controller can only handle one device if it has a short refresh depth, but more than one if it is longer.

[SilverSolder]

Why would the HP DIMM have so many more capacitors than the aftermarket one?   What are they trying to accomplish - what problem are they trying to prevent - by sprinkling capacitors everywhere, if the memory chips don't need them?

[MK14]

Why would the HP DIMM have so many more capacitors than the aftermarket one?   What are they trying to accomplish - what problem are they trying to prevent - by sprinkling capacitors everywhere, if the memory chips don't need them?

As other(s) in this thread, seem to be trying to tell you.

The capacitors seem to be a RED HERRING, and have probably got nothing to do with why only one stick works, not two.

Memory controllers, can be very fussy, and need the right type of memory. I suspect you are NOT using the right memory type (as in exact part numbers, and all applicable parameters for them), in order to use BOTH slots.

This is normal, and found in many memory related things.

tl;dr
You seem to have bought the wrong memory stick type for your device. It even has a different part number.

My guess would be you haven't bought a low enough 'rank' memory stick type.

As with PCs, they can be very fussy about what memory they have fitted to them. Especially if you want to fill out all the memory slots, as that especially needs the CORRECT memory type(s).

I.e. (Taking modern day memory sticks to make a point).
You can't just slot in a DDR4 memory device into a slot somewhere, and expect it to always work perfectly. It needs to be the CORRECT DDR4 memory type for that system. I.e. Speed and a whole lot of other parameters and/or type numbers.

[MK14]

Alternatively, you are trying to put in more memory capacity, than is allowed, in that system. Although, you can sometimes successfully, do that, in some cases.

[magic]

When you are dealing with memory subsytems, there's no such thing as "just a 128MB DIMM". The stick has an internal organization built out of the chips that are on it. Some sticks are "double ranked" or "2R" and are recognized by the memory controller as two separate memory devices. That is most likely what is going on with your DIMM.

Another possibility is that the refresh depth is different and the memory controller can only handle one device if it has a short refresh depth, but more than one if it is longer.

Having successfully fitted 0.5GB of RAM into my old Pentium II machine I can only confirm this.

Count how many RAM ICs are on each module, find their datasheets and see if the internal organization is the same. Tip: it probably is not.

[tom66]

Why would the HP DIMM have so many more capacitors than the aftermarket one?   What are they trying to accomplish - what problem are they trying to prevent - by sprinkling capacitors everywhere, if the memory chips don't need them?

Because the engineers thought they needed them.  Maybe the memory devices are older or of a lower grade and more sensitive to supply voltage. Maybe this DIMM is older and was made when high density ceramic capacitors were rare and expensive.

Some newer memory ICs now have embedded ceramic decoupling capacitors in the package in order to achieve the performance that is required - it's unlikely that this is the case for your board but it's one reason you might see more spartan memory modules nowadays.

Equivalent question - why does the newer DIMM have loads of series termination resistors on it when the older DIMM does not? Surely that is some way to cut cost too...

[SilverSolder]

The HP does have a stack of resistors on the other side, similar to the aftermarket one.

Assuming the memory chips are good, what can you possibly mess up on a DIMM other than decoupling and series resistors?

Maybe they are just using rejected chips, or they could be fake...

[Haenk]

32x32 is a strange organisation - 64x16 times 2 might work.

Funny sidenote: The aftermarket module uses used "pull"-chips (desoldered), probably without declaring so. (But new SDRAM chips would be extremely difficult to source, whereas used chips are next to "free" and plenty...)

[SilverSolder]

32x32 is a strange organisation - 64x16 times 2 might work.

Funny sidenote: The aftermarket module uses used "pull"-chips (desoldered), probably without declaring so. (But new SDRAM chips would be extremely difficult to source, whereas used chips are next to "free" and plenty...)

How can you tell - the chips are too old? ... I guess 0212 is a bit old...   -  so at least the chips are probably not fakes! 

So if the decoupling is acceptable, these chips simply aren't compatible with the motherboard, for whatever reason?

[amyk]

Like everyone else here has been trying to lead you to the right solution, you need to look up the part number on the ICs of the working one and compare the "dimensions".

[SilverSolder]

The solution was to buy the original HP sticks, used.  I guess they went through a more thorough design and test process than the aftermarket ones...

[cvanc]

Do I see a solder bridge here?

[magic]

How can you tell - the chips are too old? ... I guess 0212 is a bit old...   -  so at least the chips are probably not fakes! 

Because only one of them is 0212, that's why

[wraper]

I count what looks like 17 caps on one the HP module, plus what looks like a large tantalum cap.  I see 8 caps on the aftermarket one.  Am I missing something obvious?

You are missing that you are comparing opposite sides of the RAM sticks. There may be more capacitors on the other side.

[wraper]

Do I see a solder bridge here?

Might be, but it's a stock photo, not actual module he has.

[Haenk]

How can you tell - the chips are too old? ... I guess 0212 is a bit old...   -  so at least the chips are probably not fakes! 

So if the decoupling is acceptable, these chips simply aren't compatible with the motherboard, for whatever reason?

If you look at the second chip from the right, the "Infineon" laser marking is half readable and half "darkened". That is a 100% sign of a sticker being removed from that chip.
A slightly varying datecode was not uncommon (at that time), as each chip had a separate feeder, but in this case, they just desolder old modules, collect enough chips of the same make and model until they can do their own production run. The PCB and other parts are likely "new".

Presumably there is no way to source a larger amount of brand new SDRAM-chips.

The memory controllers were somewhat picky at that time. x8 was good for pretty much everything, x16 on higher capacity. Or x4 on AMD boards (mostly made from defective x8 chips).

[ejeffrey]

The solution was to buy the original HP sticks, used.  I guess they went through a more thorough design and test process than the aftermarket ones...

It is not design and test.  Almost certainly the aftermarket DIMM is completely meeting specifications.  It is probably exactly what everyone else has said: the number of memory ranks supported by the memory controller in the printer. 

Memory ranks are chip select lines.  Each DIMM socket has two chip select pins (well, the max # depends on the standard). In a single rank DIMM all the chips will be connected to the same chip select.  In a dual rank DIMM half of the chips are connected to one chip select and half to the other.

A memory controller like this might support either 2 or 4 chip select lines.  If it supports 4, each DIMM socket will get its own 2 chip select lines.  If it only supports two they will be shared between the two sockets but swapped on the motherboard.  That is, slot 0 chip select 0 -> slot 1 chip select 1 and vice versa. 

If you install two single-rank DIMMs, it will work fine.  Each DIMM will use just one chip select.  If you install a dual rank DIMM in one socket it will also work, using both chip select lines.  But if you install two dual-rank DIMMS on a memory control that only supports two ranks it will not work.

The exact details such as the max # of ranks and so on depend on the memory standard, but this is the main idea.

So it has *nothing* to do with how thorough the design and test procedure was and everything to do with the printers memory controller not supporting the memory configuration you used.

[SilverSolder]

The solution was to buy the original HP sticks, used.  I guess they went through a more thorough design and test process than the aftermarket ones...

It is not design and test.  Almost certainly the aftermarket DIMM is completely meeting specifications.  It is probably exactly what everyone else has said: the number of memory ranks supported by the memory controller in the printer. 

Memory ranks are chip select lines.  Each DIMM socket has two chip select pins (well, the max # depends on the standard). In a single rank DIMM all the chips will be connected to the same chip select.  In a dual rank DIMM half of the chips are connected to one chip select and half to the other.

A memory controller like this might support either 2 or 4 chip select lines.  If it supports 4, each DIMM socket will get its own 2 chip select lines.  If it only supports two they will be shared between the two sockets but swapped on the motherboard.  That is, slot 0 chip select 0 -> slot 1 chip select 1 and vice versa. 

If you install two single-rank DIMMs, it will work fine.  Each DIMM will use just one chip select.  If you install a dual rank DIMM in one socket it will also work, using both chip select lines.  But if you install two dual-rank DIMMS on a memory control that only supports two ranks it will not work.

The exact details such as the max # of ranks and so on depend on the memory standard, but this is the main idea.

So it has *nothing* to do with how thorough the design and test procedure was and everything to do with the printers memory controller not supporting the memory configuration you used.

The printer has 4 slots, one of which is populated with special stick that has some flash and 48MB RAM - it always has to be there, whatever else you do, it is the printer OS, I beleive.

The facts are: if I put two HP branded sticks in the printer, it starts and boots, and shows the expected total amount of RAM (total 304MB).

If I try the same thing with the two "non-HP" sticks, we get the problems described above - no matter which slots they are put in.


Can the brand X sticks really be called 'compatible' if they don't actually work like the originals?