Dennis said:
Yes it's a 939, an Asus A8N-SLI Premium. I guess I will start messing
around with the timing settings and some overclocking. It normally
runs on 400mzh DDR1 - anyone have thoughts about what happens if I
load it with DDR 500mzh (PC4000)? I wont be able to do this
financially, but for next time I'm just curious if this would help to
keep 4 sticks running at 400mzh instead of it dropping down to 366 or
lower like it is now.
Remember where the limitation comes from. The BIOS is coded
according to AMD recommendations. The processor contains the
memory controller. There are two 64 bit interfaces on the side
of the CPU, that drive your four memory slots. Because of all
the noisy signals on the CPU package, there are limits as to
how hard the CPU can drive the memory bus. (This is where
having the memory controller on the Northbridge can be
an advantage, because you might get away with more aggressive
drive that way.)
The "Auto" setting will compensate for bus loading, by following
the AMD recommendations. The performance might be slower than
you'd expect.
An AMD user can manually tune their computer, according to the
performance the particular processor is capable of. These are
some options.
Config Settings
2 sticks PC3200 DDR400 Command Rate 1T
4 sticks PC3200 DDR400 Command Rate 2T
DDR333 Command Rate 1T
4 sticks PC3200 (auto) DDR333 Command Rate 2T
The AMD memory design derives a memory clock from
the CPU core clock. Rates other than the canonical values
of DDR400, 333, 266, 200 are available from the processor.
Your value of DDR316 might have happened, because of a particular
divider choice. I would have expected the BIOS though,
to get closer to one of the canonical values. Maybe
its target was DDR333 but the closest divider gave
DDR316 ?
When you buy DDR500 memory, the implication is, the
memory was the limitation. There is no indication at
this point, that the memory is the problem. The memory
is rated PC3200/DDR400 and it isn't even being driven that
fast yet.
If I bought DDR500 memory, I'd run it with two sticks only,
one per channel. I don't really know how much of its speed
rating could be used on a S939 system. I don't remember
the test results for that. But that kind of memory
would be something an overclocker would use.
If you installed four sticks of DDR500, I seriously doubt
any motherboard (AMD or Intel) could get all the way to
that speed. Perhaps DDR480 might be about as good as
you'd do.
I think you can play with the BIOS settings, and use
manual settings to tune the performance. It is possible
you can get away with selecting DDR400 and Command Rate 2T,
leaving the CAS and friends at their normal values. I wouldn't
expect DDR400 and Command Rate 1T to work with a loaded
bus.
You definitely *don't* want to boot into Windows, until
all memory tuning is completed. Applying too much voltage
to the DDR, will make it run hotter than normal. And
on some motherboards. the slots are really too close
together for proper cooling. The two sticks in the middle
of a cluster of four, might not be receiving any cooling
air. The RAM I've used on my DDR systems, was "low Vdimm"
type, that didn't really like more than 2.7V. There were
some brands of RAM that could take a lot more. Even with
due care and attention, I had one enthusiast stick of
RAM fail, and I ended up buying some more.
Recommended tests are memtest86+ from memtest.org . That
one runs from a floppy or a CDROM. The main value of
memtest86+, is it comes closest to testing all the
memory locations. It doesn't do a good job of detecting
stress related problems.
The next test I like, is the Prime95 Torture Test for up
to four hours error free. It is available from mersenne.org/freesoft .
It runs under Windows or Linux. To avoid booting into Windows,
you can boot a Linux LiveCD (Ubuntu from ubuntu.com or
Knoppix from knopper.net), then use the built-in web
browser to get a Linux copy of Prime95 for testing.
A Prime95 test thread stops on the first error it
finds, and a poorly tuned system will error out within
seconds when Prime95 is running. (You can store your
downloaded copy of Prime95, on a USB stick if you want,
for reuse later.)
Once you're reasonably stable (have completed a four hour
run), you're still not done. Boot into Windows and
repeat Prime95. I've run both Prime95 and played a
3D game at the same time, and had one Prime95 thread
fail preferentially. (Set the "amount of memory to test"
a bit lower, leaving room for a game to load.) So it is
possible that the added electrical noise of running a game,
can tip Prime95 over. And it means you still have some work
to do on your RAM timing. For example, if you were running
DDR400 2T, had applied 2.7V to the DIMMs, and it still threw
errors, you might have to back off some more. To DDR333 with
some level of Command Rate. Slapping in the DDR500 RAM
might not make any appreciable difference in this
case (at least with four DIMMs).
DIMMs have two bus connections. They have an address bus
and a data bus. The address bus is the "speed limited" one.
All the memory chips form a parallel load on the address.
Selecting Command Rate 2T, presents the address data for
two cycles of the bus, and the data is strobed on the
second cycle. This gives a bit more setup time to rising
clock edge. But presenting an address for two cycles,
also cuts the address bus bandwidth in half, so fewer
addresses can be presented per second. The effect is
not directly proportional, because that bus is never 100%
busy. Using a single sided DIMM, with 8 chips, rather than
a double sided 16 chip DIMM, would be an example of a
way to unload the bus a bit, and allow higher empirical
tuning values to be used. But if you want tons of
RAM, you generally end up using the 16 chip modules.
Running just two sticks, the largest capacity DIMMs you
own, is a compromise between available memory and
operating speed. I would at least try one test that
way, and see if your subjective feelings that the
system is slow, are still present or not. And verify
with CPUZ, what settings the BIOS actually used
for the new memory by itself.
Paul
