By: Björn Ragnar Björnsson (bjorn.ragnar.delete@this.gmail.com), December 19, 2020 5:25 pm
Room: Moderated Discussions
Wes Felter (wmf.delete@this.felter.org) on December 18, 2020 9:38 pm wrote:
> Björn Ragnar Björnsson (bjorn.ragnar.delete@this.gmail.com) on December 16, 2020 10:09 pm wrote:
>
> > This in-line ECC appears to be a colossal kludge, presented as
> > a feature, to solve a problem that never should have existed.
>
> I don't think in-band ECC has much to do with Intel's price discrimination around ECC. On a DDR4
> memory channel you add a 9th DRAM chip for ECC. On an LPDDR4 channel based on x16 or x32 DRAMs...
> you can't just add another chip. I guess there's a parallel universe where LPDDR4 was defined
> as 36-bit or 40-bit wide but it's too late for that now. I also suspect that if Intel had never
> skimped on ECC some other company would have eventually started the race to the bottom.
There's a legitimate need for ECC for systems that have no way of providing support for it. I suspect that something like socket compatibility is driving this in-band ECC push. If you don't have the pins for ECC bits in a socket you now need to support with ECC, I guess in-band ECC is reasonable far from optimal surgery.
Race to the bottom by anybody other than Intel would have failed and the reason is simple: Intel has for the ~ last three decades held nearly all of the CPU market power.
My take, ECC support should never have been dropped the way it was and the way it was dropped constitutes in my mind malicious price discrimination. I realize that price discrimination is probably inevitable in many circumstance, I just wish Intel would have discriminated less destructively.
> Björn Ragnar Björnsson (bjorn.ragnar.delete@this.gmail.com) on December 16, 2020 10:09 pm wrote:
>
> > This in-line ECC appears to be a colossal kludge, presented as
> > a feature, to solve a problem that never should have existed.
>
> I don't think in-band ECC has much to do with Intel's price discrimination around ECC. On a DDR4
> memory channel you add a 9th DRAM chip for ECC. On an LPDDR4 channel based on x16 or x32 DRAMs...
> you can't just add another chip. I guess there's a parallel universe where LPDDR4 was defined
> as 36-bit or 40-bit wide but it's too late for that now. I also suspect that if Intel had never
> skimped on ECC some other company would have eventually started the race to the bottom.
There's a legitimate need for ECC for systems that have no way of providing support for it. I suspect that something like socket compatibility is driving this in-band ECC push. If you don't have the pins for ECC bits in a socket you now need to support with ECC, I guess in-band ECC is reasonable far from optimal surgery.
Race to the bottom by anybody other than Intel would have failed and the reason is simple: Intel has for the ~ last three decades held nearly all of the CPU market power.
My take, ECC support should never have been dropped the way it was and the way it was dropped constitutes in my mind malicious price discrimination. I realize that price discrimination is probably inevitable in many circumstance, I just wish Intel would have discriminated less destructively.
