What board and chipset are you running? That is the biggest determinator of what memory should work, not just what fits in the hole. Meaning you could have a pile of working memory still, just not working for your board. Density is usually the biggest problem, but there's other things like tight timings given in the SPD that the chipset can't cope with, especially if it's on the bottom side of voltage spec. There's a sort of hierarchy of density and other factors, so if it doesn't work in a 430VX or a 440LX, try it in a 440BX or Apollo, if it doesn't work in those, try it in a i815 or Apollo Pro, if it don't work in those, a KT133 or P4 SiS SDRAM board.. and if it doesn't work in any of those desktop chipsets, maybe it's a weirdy server module or special purpose.

All of these were tested with an Intel 440EX with a Pentium 2.

These modules are not the issue. These timing is not the issue as you have OEM brand modules which means quality.

Look at your motherboard, post some photos of your motherboard in good detail, and how is the CPU's heatsink and good thermal paste applied?

EX chipset is crippled and uses low density modules meaning 128MB as 8 or 9 chips total or 256MB with 16 or 18 chips total each.

Cheers,

Is chip density a visible as a field in EVEREST Home Edition? For the chips that are unstable, nothing stands out as different compared to the others that work fine.

pentiumspeed wrote on 2021-08-19, 14:30:

These modules are not the issue. These timing is not the issue as you have OEM brand modules which means quality. […]
Show full quote

These modules are not the issue. These timing is not the issue as you have OEM brand modules which means quality.

Look at your motherboard, post some photos of your motherboard in good detail, and how is the CPU's heatsink and good thermal paste applied?

EX chipset is crippled and uses low density modules meaning 128MB as 8 or 9 chips total or 256MB with 16 or 18 chips total each.

Cheers,

"Low density" isn't really a helpful term, as what is "low" or "high" changes with time. Much better to just state the density itself, that's unambiguous.

i440EX officially supports up t0 64Mb memory density, so 64MB with single-sided 8/9 chip DIMMs or 128MB with double-sided 16/18 chip DIMMs.
However the i440LX and EX will also happily work with 128Mb density chips, giving you twice that, the sizes you are referring to. When the i440EX was current this was considered very high density, it only became low density when the 800-series chipsets and later Via ApolloPro133A and KT133(A) started supporting 256Mb chips.

Also, I disagree that the symptoms are caused by memory density issues. If it's just the density that's too high, the DIMM will work, but only be usable at the highest capacity the memory controller can handle. So a DIMM with 256Mb chips would show up as 128/256MB instead of 256/512MB. It would not cause Memtest to find errors or the system completely fail to boot. With a 66MHz bus speed, low speed spec isn't going to be a problem either.
The only specification issue that would cause a boot failure would be x4 chips (i.e. 128Mx4 64Mb chips or 256Mx4 128Mb chips), as JEDEC spec and indeed Intel memory controllers do not support DIMMs with unbuffered x4 memory chips. DIMMs with this kind of layout were a common pain around the turn of the millennium, frequently marketed as "Via only"memory as Via's memory controllers were happy to use them.

However, I agree that all relevant compatibility data should be given here, so:
- motherboard vendor and model (we already know it has an Intel 440EX chipset)
- per DIMM its vendor and model, if known
- per DIMM the SDRAM chip code

Kahenraz wrote on 2021-08-19, 13:07:

None of the defective sticks had a name brand etched onto their chips such as Hyundai, Hynix, Micron, or Samsung.

That's probably a good hint . Re-marking probably wasn't yet a big thing back then but selling modules with marginal, usually unbranded, chips (that likely failed QA at whoever made them) was a thing .

darry wrote on 2021-08-19, 23:53:

Kahenraz wrote on 2021-08-19, 13:07:

None of the defective sticks had a name brand etched onto their chips such as Hyundai, Hynix, Micron, or Samsung.

That's probably a good hint . Re-marking probably wasn't yet a big thing back then but selling modules with marginal, usually unbranded, chips (that likely failed QA at whoever made them) was a thing .

This is my gut reaction as well. The only 100% dead stick actually had almost no markings at all. Very suspicious.

There are a couple numbers on the chips: 1608-210014. That's it.