It always amazes me how people can be so confident yet so wrong.
It's a problem of physics - there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM and that is going to be extremely expensive when talking 16GB and not nearly dense enough.
It's not some conspiracy by "Big RAM"
And yet, when the bit flip problem caused by physics got so bad in DDR5 it couldn't be ignored they did fix it - by adding error correction codes. It wasn't that expensive to do so. Notice that HDD's hit the same problem as the got denser, and solved it in the same way (ie, by throwing lots of ECC at it).
I agree with the original poster. It's a hardware problem, caused by the manufacturers pushing the limits. And it's their problem to fix, which they can do. DRAM that doesn't corrupt itself isn't a big ask.
I mean, if you had a DRAM labeled "DDR4-3200" but it can only work at a much lower speed (say DDR4-2400), it's clearly a problem of physics - the gate capacitance is too high, the driver transistors are not strong enough. And yet my reaction would be to take that RAM back to store and get my money back, not to defend manufacturers which claim false things about their chips.
Why wouldn't storing a cryptographically-secure checksum on every RAM row work?
Yes it's a problem of physics and it's because they are trying to make DRAM too dense.
It's a problem of physics - there's various ways to try to mitigate it but the only way to completely avoid it would probably be to use SRAM
It's not some conspiracy by "Big RAM"
Look at the evidence. This didn't start showing up until around 2009-2010, and the industry managed to convince authors of widely-used memory testing tools to downplay the severity and/or not enable RH tests by default, because they didn't want the truth to be known that almost all RAM is defective. It might not be a conspiracy, but it sure is corporate greed.
Would you rather pay a little more for RAM that will work correctly under all access patterns, or RAM that is certain to produce bit errors under some conditions that can be encountered in practice? Unfortunately, with newer DDR3 and later generations, it seems you don't get a choice.
The choice just boils down to "buy pricey server boards/chipsets that have ECC RAM available" or "get bent".
ECC still superior by a longshot, this information notwithstanding.