Post by JackPost by PaulYou could use one type of RAM,
or the other type of RAM, but not mix them.
How about mixing DDR4 3200MHz with DDR4 3600 MHz? Is that going to pose
any problems?
I have 3200MHz ram from Corsair (2 x 16 = 32 GB) and I intend to stock
up with 2 more sticks (2 x 16) but they only have 3600 MHz in Vengeance
LPX model.
My machine can take up to 128 GB ram but I thought 64 GB is more than
enough for most things. In fact even that is an overkill but DDR4 is
reaching EOL so I thought I better get it. The price is not bad on e-shops.
The brand I have is CMK32GX4M2D3200C16 but amazon has brand
CMK32GX4M2D3600C18
You can mix them.
The BIOS does the math, to convert between clock speeds and
compute a new CAS value. The "slowest" CAS value is then used
for all the sticks. The intention is, if you mix a set of
sticks, they select conditions that work for all of them.
The controller, can only run one set of conditions, for all
the installed sticks.
Where you can get caught, is if one DIMM uses 2.5V and
a second DIMM uses 2.7V and a third uses 2.75V . The BIOS
does not consider the voltage, and while the DIMMs have
timing specs, for "enthusiast speeds" the tighter timings
are only available at the higher voltage. Well, on some
motherboards, you don't get to adjust the voltage. Back in those
days, there was no such thing as XMP (XMP being timing+voltage spec).
But for conventional RAM (stuff close to JEDEC specs),
the system "just works". On my refurb computer, nothing
bad can ever happen, and "good RAM" runs at "bad rates" :-)
But, if you're not paying attention, it all works.
Dells and HPs are meant to be "Plug and Play", even if
the performance isn't at Max. The Dells and HPs would
use columns on the left in the following picture, rather
than absolutely guaranteeing to use the right-most column.
If you are mixing enthusiast sticks, then how bad it
can become, depends on which generation of RAM is involved.
[Picture]
Loading Image...If you mix sticks, I don't know if switching on XMP will work
or not. With a retail motherboard, you should be able to
manually set the voltage to 1.35V . Since both the 3200 and
the 3600 sticks will be running at 1.35V, then the usual maths
by the BIOS, will pick the lowest common denominator.
The main benefit of matching all the sticks (from that CPUZ picture
point of view), is improving the odds that good things will happen.
I can tell you from experience, there is a HUGE difference
between manually adjusting RAM, versus "hitting XMP and done".
On this system, I switched on XMP (even though it's not supposed
to work with four matched sticks, and it is specced to quantify two sticks),
and everything "just worked". Which is a minor miracle compared
to other RAMs I've used. The DDR3 in the other machine was
pure misery (Intel, bus loading) to deal with. Filling all the slots
with RAM on that one, was a mistake. It took around one week of testing,
before I could close up the BIOS and enjoy the computer.
Check the SPD table on yours. Put just one stick of the 3200, do a CPUZ.
Put just one stick of the 3600, do a CPUZ. Then puzzle over the table
and consider what a mess you're in. That's how I'd have to do it.
But I assume both the 3200 and the 3600 will look like my table,
and they're actually using more voltage to reach the right-most column
of numbers. (And that column is where everyone runs them. The
other columns only exist, to get a motherboard started after
you finish assembly and turn on the power. The other columns
are the "guaranteed to start" columns. The XMP column might not work,
then if you "reset" the BIOS, you can start over again with the settings.)
How XMP *should* work, is it should only allow you to switch on XMP,
if "sufficient matching is present". If sufficient matching is
not present, you'd set the voltage to 1.35 and select a manual
value. (For example, you would set the clock rate to 1600 for the
RAM, which is DDR4-3200, leave all other settings at defaults
so it can do the math for you, and then try and run Memtest.)
The combination of giving it 1.35V, plus telling it to run
at 1600MHz, should get you very close on a modern system.
You run Memtest then, see if you're getting
No errors.
Just a couple errors over an entire Pass.
Buckets of errors (indicating a foul-up somewhere in your choices).
The most errors I ever had, was a stick of Crucial Ballistix, where
one chip just up and died. And basically, every read was filled
with errors, and the thing was just scrolling like crazy :-)
That constitutes a "more than Buckets" rating. A Deluge of errors.
On one of my motherboards, I changed the FSB setting one day.
And a little later, I ran CPUZ for fun, and the bloody BIOS
had not adjusted the RAM, and the RAM was being overclocked by
something like 33%. Well, this is how you learn which things
are "independent" or are "clock-locked" :-) If it wasn't for my
luck, it would have crashed, and I would have assumed my CPU
was "lame". BIOS have their quirks, and on some systems,
the owners figure out just by the boiling hot temperatures
of the equipment, that the wrong voltage is being applied to the
CPU.
Modern hardware (what you're using), has several clock domains,
and when I tell you to set the clock to 1600, I can do that
because I know the control only adjust the RAM, and does not
fiddle other stuff. Nothing else should go bonkers when you
set the memory speed in the BIOS. The memory speed and the
"gear setting" on the memory controller, are the things
that are linked. If you select a too high clock for memory,
the gear selector can go from 1:1, to 2:1 or 4:1, and the
places where overclocking is "beneficial" is only over
limited ranges of the frequency domain.
I'm running 3200 RAM on a system that can run 1:1 at 3600.
And I did that, so if the RAM needs to be sold on some
day, the RAM can be used by anybody. I didn't want to buy
RAM that was only good for one kind of system. Like, when you
flip on the XMP, your prospective customer will really like
the purchase, if the XMP switch "just works" and memtest passes.
And the 3200 passed no problem, on my system, even though I
was enabling XMP, while using four matched sticks (not two
sticks like the table assumes).
Paul