Norm X wrote:
"VanguardLH" wrote in message
...
Norm X wrote:

Happy New Year,

My old Intel MOBO has accepted many upgrades but is stuck at 4GB RAM.
Even
though prohibited, Win10 works reliably on one stick SOCHOX 4GB 667MHz
DDR2
AMD, the past couple of years. However, when two such modules are plugged
in, Win10 boots until it panics and flashes a QR code error and shuts
down.
Paul has attributed such failure to power supply inadequacies of Intel
MOBO
versus AMD.

Still wanting the best that can be achieved with least money, I came
across
"SOCHOX 8GB DDR2 PC2-6400 800Mhz DIMM AMD " on eBay. It comes in a two
module 16GB package, while I only need 8GB. At first blush, 800MHz seems
better than 667MHz. Current CPU-Z output says this newer RAM module may
be a
good idea.

Comments?
The reason there is a limit on the total capacity of system memory on
the mobo (besides power distribution constraints) is there just aren't
the necessary address lines to beyond that maximum addressable range.
The 4 GB limit means 32 address lines (2^32 = 4 GB) on the mobo. If the
mobo doesn't physically have more address lines (which go to each
module/slot for interleaving) then you cannot get beyond 4 GB. When you
add the 2nd module, and due to interleaving or dual mode or both, you're
trying to use more capacity than the mobo can handle.

If the mobo says its addressing limit is 4 GB, that means the total
across all slots. Doesn't matter if you use only 1 slot or all 2 or 4
slots. Yeah, it's sad when you see the unused slots and want to fill
them but if you've maxed out the memory addressable range of the mobo
then you didn't plan well back when you bought the mobo for the
foreseeable future of owning that mobo. If you want more system RAM,
you need to get a mobo that specs out to handle it.

Also, despite going higher on the memory module's clock rate, there is
no way the mobo is going to magically handle clock rates higher than
what its own clock can handle. You can get a bigger bucket but it's
still going to get filled only as much as before with the smaller
bucket. Only if you plan on overclocking the mobo and hope it, the CPU,
and memory are stable at the higher clock rate does it make sense to get
the higher clockable memory (because you then aren't overclocking that
memory but are overclocking the CPU -- unless the mobo gives separate
clock dividers for CPU and memory but the mismatch means the CPU has to
wait to get in sync so you don't gain much, if anything, and could
actually slow the PC).

If you want more system RAM than what your current mobo can handle at
its designed max of 4 GB, you need a new mobo. If you overclock,
getting higher clockable memory means one less component to go unstable.
That's like one sane guy in a house of crazies. If you don't overclock
then only bother with higher clockable memory modules if they are priced
the same or sometimes cheaper than the max clock modules the mobo
specifies (yep, sometimes the higher clockable modules are cheaper then
the slower modules). A $1.2M Hennessey Venom GT that goes 0 to 149 and
back to 0 in 30 seconds but stuck in rush-hour traffic is a sad joke.

To get more system RAM means a new mobo. A new mobo also means you can
go to higher clockable DDR4 memory modules.

Thanks for the comment. I am using an Intel Q6600 CPU. See:
https://ark.intel.com/products/29765...z-1066-MHz-FSB

"Intel® 64 architecture delivers 64-bit computing on server, workstation,
desktop and mobile platforms when combined with supporting software.¹ Intel
64 architecture improves performance by allowing systems to address more
than 4 GB of both virtual and physical memory."

I think that the easy explanation of how this works, is by multiplexing.
E.g. USB is more versatile than parallel I/O. Within a CPU, all address and
data lines may be present, but outside the package it is multiplexed. The
Q6600 is LGA775, but up to LGA 2066 exists. LGA2066 may be super-duper but
is can only be super-expensive.

Actually, you can get "cheap" LGA2066 processors.

https://ark.intel.com/products/12149...up-to-4_50-GHz

The good part:

# of Cores 4
# of Threads 8
Processor Base Frequency 4.30 GHz

The bad part:

Only two of four memory channels work. That means four of eight
DIMM slots are "dead". Nothing is wired to them, in effect.

Only 16 PCI Express lanes exist on the processor
(LGA2066 CPUs come with 16, 28, or 44 lanes, and the
motherboard manual tells you which slots stop working
if you buy the "puny" stuff.)

CPU is missing AVX512 apparently, compared to the high end ones.

It is little better than a 4790 in a way. If they made
a MicroATX motherboard with just four memory slots, and
one video card slot, plus the LGA2066 socket, that processor
would be OK. But mixing an "expensive" motherboard with that
piece of crap, is just silly. It's just a way for Intel to make
money for nothing (charge more for chipset).

But the mis-appropriation of features, makes the socket
choice a terrible one. It's just an LGA1151 CPU on a LGA2066
substrate.

Also, some desktop motherboards take Xeon branded CPUs, and the
CPU Support chart shows official motherboard company support for
them. If you see a bargain somewhere, Xeons offer an alternative
means of getting a processor for your motherboard.

*******

On some chipsets, the actual chipset had a "max addressable".
My chipset, I think it won't go above 8GB. So even if I found
DDR2 modules that were addressable, it wouldn't help.

There was one Intel chipset, where the slots took 4x2GB,
yet the chipset had a defined 4GB limit. Practical configurations
ended up being 4x1GB or 2x2GB (with two slots empty).

The absolute worst, was the Intel chipset that supported
512MB DIMMs per slot, had three slots, yet the entire chipset
had a *512MB* limit. Useful configurations might be
256+128+128 all the way to 512MB,nothing,nothing. Now,
how is that for wasting slots ?

I have a TUV4X from that era, a VIA board, and it took
3x512MB and that's what I had in it when I tested it
six months ago. It was a spare motherboard that I'd never
used. The thing worked fine. When VIA did stuff like that,
that's why Intel had to get even with them, and take their
license away. And push them out of the chipset business.
Now, Intel has all the chipsets (PCH) to itself.

Paul