Re: [gentoo-user] Re: Package compile failures with "internal compiler error: Segmentation fault".

[Michael <confabulate@kintzios.com>]

[-- Attachment #1: Type: text/plain, Size: 1480 bytes --]

On Thursday 5 September 2024 10:08:08 BST Frank Steinmetzger wrote:
> Am Wed, Sep 04, 2024 at 11:38:01PM +0100 schrieb Michael:
> > Some MoBos are more tolerant than others.
> > 
> > Regarding Dale's question, which has already been answered - yes, anything
> > the bad memory has touched is suspect of corruption.  Without ECC RAM a
> > dodgy module can cause a lot of damage before it is discovered.
> 
> Actually I was wondering: DDR5 has built-in ECC. But that’s not the same as
> the server-grade stuff, because it all happens inside the module with no
> communication to the CPU or the OS. So what is the point of it if it still
> causes errors like in Dale’s case?
> 
> Maybe that it only catches 1-bit errors, but Dale has more broken bits?

Or it could be Dale's kit is DDR4?

Either way, as you say DDR5 is manufactured with On-Die ECC capable of 
correcting a single-bit error, necessary because DDR5 chip density has 
increased to the point where single-bit flip errors become unavoidable.  It 
also allows manufacturers to ship chips which would otherwise fail the JEDEC 
specification.  On-Die ECC will only correct bit flips *within* the memory 
chip.

Conventional Side-Band ECC with one additional chip dedicated to ECC 
correction is capable of correcting errors while data is being moved by the 
memory controller between the memory module and CPU/GPU.  It performs much 
more heavy lifting and this is why ECC memory is slower.

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]