On Thu, 3/12/2026 10:19 AM, John W. wrote:
I have 4 GB 1333 MHz DDR3 RAM installed in socket 1 ((of 4) on my motherboard,
and 4 GB in socket 2 as recommended by Gigabyte for Dual Channel mode.
I want to increase my RAM to 16 GB. I could add 4 GB in socket 3 and 4 GB in socket 4, or a single 8 GB module in socket 3. Would it make any difference apart from a single 8 GB module being $10 cheaper?
I can't promise anything, but RAM lore works better if we know
CPU socket, CPU model number, maybe chipset (X38, Q45, E870).
Convenient lookup on sites like crucial.com is gone. A crucial.com
entry as a "hint", that could hint at what the root cause of a
RAM screwup was.
Older designs had
1) Limited memory map (16GB max, no matter how big a DIMM was)
2) Memory controller in Northbridge (chipset), like X38
Newer designs had
1) Memory map limit unknown (nobody bumps head while testing, all RAM registers)
2) Memory controller and memory I/O pins moved to CPU package.
3) Still pretty severe loading effect with 2 DPC (dimms per channel)
AMD used 8mA drive to avoid SSO.
4) First gens of AMD MemCtrl in CPU, were true dual channel and
it was too much work to design Flex Memory into the memory controller.
We carefully matched pairs of DIMMs for those (DIMMs accessed in
"128 bit mode").
As memory standards change, the speed of the arrays inside changes too.
There can be as many as eight planes inside a memory chip (memory chip is
not stacked and this is just an architecture statement). In some cases,
the bus loading effect had just disappeared (my DDR4 systems 1DPC or 2DPC
runs at exactly same speed and CAS timing, DDR5 are back to bus load
effects again).
To answer the question from 60,000 feet, yes, (4) 4GB allows an alternating fetch from the stick pairs. The stick pairs then get a benefit from
"double the number of open pages", this is a 1% improvement for the CPU, depending on the "stride" of program access. This may have been
termed "Symmetric" on one Intel generation.
Ch0 Ch1 Ch0 Ch1
| | | |
4GB SS 4GB SS --- ---
| | | |
4GB SS 4GB SS 8GB DS 8GB DS (higher memory clock)
(watch for density limits, some CPUs lack docx)
Ch0 Ch1
| |
4GB SS ---
| |
4GB SS 8GB DS (BIOS could use lower clock, config becomes single channel
if the arch is not a Flex Memory one)
For pre-built machines (a Dell BIOS), the BIOS will select the most
shitty memory settings it can think of. Enthusiast motherboards
are more tune-able, even if it takes a week of test to determine
that the user tuning is error free. As a result, we do not speculate
on getting a good "treatment" from a canned-BIOS. My Dell DDR3 has (4) x 4GB DDR3
sticks and runs them slow anyway. No complaints, but also no "wow" factor.
On a number of occasions, I have had to root cause memory density
problems ("high density" versus "low density" RAM is a simple
mental concept versus explaining to people some address bus bit
needed is missing). On one occasion, the SPD on the DIMM was
the wrong one, and belonged on some other DIMM. The BIOS is pretty
clever and that computer did not crash! Kingston was shipping
mixed DIMM solutions, and mixing high and low density. You could
buy two packages, one happened to be high density, the other
low density. This is a contract-manufacturing problem (not enough
whip and chair when dealing with idiots).
Paul
--- PyGate Linux v1.5.12
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)