[coreboot] 8 GB DIMMs on Nehalem (Arrandale)

[Stefan Tauner]

On Sat, 28 Jan 2017 17:01:09 +0300
Andrey Korolyov <andrey at xdel.ru> wrote:

> > The chipset in the (QC version of the) W510 is actually exactly the same as
> > in the X201 and T410s: Ibex Peak.
> >  
> 
> But CPUs we are looking at *are* actually different

Of course - I did not bring up chipsets ;) I am also not implying that
it should work on Arrandale but I'd like to know why it doesn't and the
usual "it doesnt support it" as seen around the internet (and
apparently even here) does not cut it :)

> - scale-down could
> mean an exposure of a previously unaccounted design issue which
> actually prevented 32nm CPU 'upgrade' to work reliably with >8G of
> RAM... And AFAIR nobody ever reported to break official memory limits
> within an entire family, I`ll be happy to know that this is not true.

Not sure if I interpret "within an entire family" correctly, but the
online specs for the 820QM are clearly wrong:
https://ark.intel.com/products/43124/Intel-Core-i7-820QM-Processor-8M-Cache-1_73-GHz

The datasheet (320765) is a bit more correct because it leaves room
for interpretation:

 * Using 2-Gb device technologies, the largest memory capacity possible
   is 8 GB, assuming dual-channel mode with two x8, double-sided,
   un-buffered, non-ECC, SO-DIMM memory configuration."
 * Up to 32 simultaneous open pages, 16 per channel (assuming 4 Ranks
   of 8 Bank Devices)

The ICs on my 8 GB DIMMs are 4 Gb (Hynix H5TC4G83BFR and whatever
Kingston uses on their KVR16LS11/8 DIMMs but clearly specified as 4 Gb
chips).

Now... guess what the datasheet of corresponding to the i5-520m (322812)
states. Exactly the same as above. I can see some other memory-related
differences (e.g., regarding support for 1333 MT/s that's only
available on the i7) though but nothing related to memory organization,
addressing or the like (but the datasheets are very superficial
regarding the memory controller as all corebooters know).

So of course the CPUs are not equal and of course the manufacturing
process or small changes in the design *could* make the difference but
it is all but clear from the public documentation what the true maximum
is - at least for me. And I don't yet accept that it is an
architectural/digital design problem.

> Just to add .02c - DDR3 interposers from tek/futureplus are appearing
> on ebay relatively frequently and for moderate price, so if someone
> has will to spend some effort on this task, the analyzer coupled with
> interposer could provide great help, with regard to unstable behavior
> with a single 8G DIMM.

Definitely not worth it for this 7+ year old hardware (unless someone
has a suitable scope and the knowledge already... the best scope I have
access to is an Agilent MSO6104A but no idea where to even start - and
no interposer obviously) but you seem to imply that the 8GB problem is
an analogous one... do you think about the 16 GB problem as well?

-- 
Kind regards/Mit freundlichen Grüßen, Stefan Tauner