The said:
I have the Gigabyte GA-880GA-UD3H motherboard and I'm looking to buy
some RAM for it. Will this board support DDR3 1600 RAM? The compatible
RAM list from their site says it will, but the specs that came with
the MB don't mention it. The manual just says ". . . Support for DDR3
1866(OC)/1333/1066 MHz memory modules". So I'm kinda in a quandary
about this. Can anyone shed some light please?? Thank you.
So you're wondering whether intermediate values are possible ?
You can start by looking at the user manual for your motherboard.
The RAM timing section indicates it has four settings (dividers).
Combined with that, would be doing things like changing the CPU
input clock. If you bump up the CPU input clock, bump down the
CPU multiplier, bump down the HT multiplier (to remain within
spec), then memory values above stock should be possible. That's
how you manage to get over 1600. I think the AMD stock may be
1333, with the possibility that four sticks runs slower than
two sticks (I'd verify this for you, if AMD still had any kind
of data sheet). Maybe if you install four sticks, you run it
at 1066 ?
RAM Divider "DDR Rate" (memory clock value as seen in CPUZ)
Nominal x4.00 DDR3-800 (400)
CPU Input x5.33 DDR3-1066 (533)
= 200MHz x6.66 DDR3-1333 (667)
x8.00 DDR3-1600 (800)
Faster RAM will also run at slower speeds, no problem. So imagine
that you're not happy with the conditions to run at DDR3-1600
(the rated speed of the RAM). You can always leave the processor
at stock, and select DDR3-1333 in the BIOS screen, and run the
DIMM at 1333. It will run fine below the rated clock. So a
fast DIMM, like DDR3-2000, could be used for overclocking
experiments, as well as running at AMD stock of DDR3-1333.
The timings (tRAS, tRCD, tCAS and the like), are measured in
"ticks". Like CAS7, means 7 clock cycles. The clock cycle in
that case, is the speed the DIMM is running at (800MHz on the
DDR3-1600 DIMM). The clock period of an 800MHz clock
is 1.25nanoseconds. So 7 * 1.25ns = 8.75 nanoseconds.
The timing parameter in question, might be a "minimum allowed
time". Now, when you change the memory clock (either above or
below its stock value), that changes the clock period. Then
you have to re-solve the equation: ??? * 1.5ns = 8.75 nanoseconds.
In other words, the timing values, have to be compensated for
the clock being used. The DIMM itself has a timing table stored
in the SPD. It might say DDR3-1600 CAS9, DDR3-1333 CAS8 or whatever.
If you're going to be using values which are not in the internal
table of the DIMM, or setting everything up manually, it's your
responsibility to scale the timing numbers (the 7-7-7-21-whatever)
thing, according to the clock used. If you don't make the adjustment,
the computer could crash or hang in the BIOS or sooner. Clearing
CMOS will allow you to recover from the blunder (some motherboards
recover on their own, and reset the BIOS after a crash, but
the CMOS jumper covers the case of really bad crashes).
Always unplug the computer, when using the clear CMOS jumper. With
any luck, the manual will warn you about this.
"Always turn off your computer and unplug the power cord
from the power outlet before clearing the CMOS values."
It helps, if you're going to be doing manual settings, to find
an article on the web that describes the "clock architecture" of
your processor. That is necessary, so you can understand, when you're
changing the CPU input clock, what parts of the processor are also
overclocked (whether you wanted it or not). The latest AMD processors,
have CPU core clock, uncore clock (internal Northbridge/L3 cache),
and memory clock. It pays to understand how they're related, whether
they change in a dependent or independent fashion or not. While I
found one site that attempted to explain it, they seemed to be missing
good info on the uncore.
The AMD processors now have more than one power plane. When the uncore
thing was introduced, the Vcore circuit around the processor was
split in two pieces. You might see five phases, but the circuit
configuration was "4+1". That means 4 phases powered the CPU core,
1 phase powered the uncore and L3 cache. A separate circuit, near the
DIMMs, might provide DIMM power (and also power the bus drivers on the
processor memory interface). So again, it helps to understand how
the power is generated, whether you can adjust the uncore voltage
separately and so on. Apparently, if you bump the memory clock,
it may be profitable from a performance perspective, to change
the uncore clock and operating voltage. It's actually a bit complicated.
But if you want to go "non-stock", you should at least understand
a little of it, so you'll understand why the computer is
crashing.
Some BIOS, when you use their built-in overclocking functions, will
do the hard work, like making minor adjustments to the uncore for
you. But at the same time, some of these BIOS have also been known
to do nutty things. Which is why, the sentient operator always has
to be on the lookout for surprises
If I'm reading the manual right, 1600 should be easy to get, and
1866 may require more research to do it right.
Good luck,
Paul
