Dual-channel mismatch sensitivity?

[Craig White]

On Fri, 2007-02-16 at 17:06 -0700, Alan Dayley wrote:

> Questions:
> 
> 1. Is dual-channel operation so sensitive that one should only run it
> with memory bought as pairs?
----
buying in pairs is your only assurance that the timing of the memory
that is to be interleaved actually matches. As Joseph pointed out (and I
completely failed to interject), non-ECC memory has no means to detect
the memory errors so any failure in interleaved pairs is left solely
with the OS to deal with and the OS doesn't have any graceful ways to
deal with it.
----
> 
> 2. Does dual-channel operation enhance performance enough to be worth it?
----
That of course is a judgment call. The only correct answer that I can
give is Lyle Lovett's analysis...




If Ford is to Chevrolet
What Dodge is to Chrysler
What Corn Flakes are to Post
Toasties
What the clear blue sky is to the
deep blue sea
What Hank Williams is to Neil
Armstrong
Can you doubt we were made for each
other?


OK...it's a pickup line but it's one of my favorite Lyle songs.
----
> Discussion:
> 
> In the thread I started on how to determine in-system RAM
> specifications, Joseph pointed out:
> 
> "Unpaired DIMM's will still work, the dual-channel interleave just
> won't. You may have to set your BIOS to not use dual-channel to make it
> work properly.
> 
> The bigger issue is that, even from the same manufacturer, if the specs
> on 2 banks are slightly off (this often shows up in the non-speced
> timings), the MB may run fine in dual-channel mode, except for
> occasional errors that won't be noticed without ECC enabled, and may
> lead to random system instability.  That's why it's generally best to
> buy paired DIMM's if you want to run dual-channel RAM. If you aren't
> doing that, then just make sure dual-channel is turned off, run DIMM's
> with the same specs for flexibility, and don't worry about manufacturer
> so much."
> 
> That sounds like proper dual-channel mode operation is a pretty
> sensitive thing.  It also sounds like one could buy a RAM module today
> and another of the same part number and manufacturer 6 months from now.
>  And the second module would most certainly be of a different lot or
> production run which could cause the problem Joseph describes.
> 
> So, I see a few options for me in my situation and would like feedback
> from those with more experience with RAM concerns than I.
> 
> 1 - Buy a proper speed and size module from where ever, disable
> dual-channel and be happy with double the RAM space.  This is the safe
> and least expensive route but no dual-channel.
> 
> 2 - Buy a proper speed and size module from where ever, run dual-channel
> and risk the sensitivity.  This is risky based on Joseph's description.
>  Silent data corruption is very bad!
> 
> 3 - Buy a proper speed and size module from Toshiba (my computer
> manufacturer) and more expensive than third party RAM, run dual-channel
> and still risk the sensitivity.  This is the safest for dual-channel
> without buying double the RAM.
> 
> 4 - Buy a new RAM pair from where ever, remove the RAM I already have,
> run dual-channel with the new pair and less worry about sensitivity.
> Sell the original module on eBay.  This is the solution for confident
> dual-channel but is considerably more expensive.  (And problematic since
> I could not see the current RAM module on my first look into the case.)
> 
> Maybe I am making a mountain out of a molehill here.  If dual-channel
> performance does not make that much difference, it doesn't really matter
> and I'll just take option #1.
----
and to continue along the path of obscure metaphors...Life is like a box
of chocolates, you never know what you're gonna get. I am sure that even
buying the memory from Toshiba is no guarantee that the RAM will match
to the existing RAM because it is likely that the RAM they have in stock
has nothing to do with the RAM used to assemble the laptop in Japan or
wherever they assembled the laptop who knows how long ago.

Craig

PS...try option 1 and see what happens