Ant said:
OK. What do I change to overclock safely in CMOS? I am using an MSI K8N
NEO4-F (MS-7125; PCB v3.0) motherboard (NVIDIA nForce4). I better be
able to revert back if something goes wrong without opening up my
computer case to reset CMOS because I am physically handicapped (can't
even move and open a mini-tower case). I know I changed a CMOS setting
once, but managed to make the motherboard not boot up anymore unless
CMOS was resetted which I couldn't do (had to call up a computer friend
to do it).
Part of your prep work, to overclock, is learning how far the thing
can be pushed (based on other people's results). Also, determining whether
the motherboard in question, is overclocking friendly (recovers
gracefully if pushed too far).
For example, I overclocked my Asus Nforce2 board, and it only got stuck
just the once and needed to have the CMOS reset. The rest of the time,
the procedure was to turn off the power, and it recovered on power up.
To overclock an Athlon64 with DDR memory, you'd need to:
1) Prevent the Hypertransport bus from working outside its spec.
Normally, the master clock is 200MHz. The bus multiplier might be 5X.
200 x 5 = 1000MHz, and that is the limit for the Hypertransport clock.
To fix it, you enter the BIOS, and drop the multiplier to 4X. Then,
at 200MHz, you get 200 x 4 = 800MHz (i.e. less than the max). As you
raise the master clock, and overclock, you get closer to 1000MHz again.
The limit would be 250 x 4 = 1000MHz, so by selecting 4X, that leaves room
for up to a 25% overclock.
2) Memory can be the same way. Some BIOS show the actual memory clock and
others may just show the "nominal" memory clock. Say the memory is normally
set to run at DDR400, when the master clock is 200MHz. If you raised the
master clock by 25%, the memory would be running at DDR500, which is kinda
high. To stop that, before the overclocking experiment, you enter the BIOS
and drop the memory to DDR333. Then, if the master clock is raised by 25%,
the memory would end up at 333 * 1.25 = 416, which is still too high. If
you stopped at the 20% mark, and did no more than 20% overclocking,
333 * 1.20 = 400, which is spot on, and thus you won't be stressing the
memory.
To give an example, say your memory runs at DDR400, and the timings are
3-3-3-8. You set the memory to DDR333, and manually set the timings to
3-3-3-8. If you leave the timings at "Auto", the BIOS will recalculate
new numbers, tighter than 3-3-3-8, and you don't want that. If you use
CPUZ, you can record the information for the current settings, and
apply that manually back into the BIOS. Tras, Trcd, Trp, and so on.
Now, you're done with the BIOS. Reboot into Windows, with your 200MHz nominal
master clock to the CPU intact, and no actual overclocking having taken place.
The above two operations, slackened off the HT bus and the memory, and should
not have frozen the machine, or affected your ability to boot.
Now, run your benchmark again, and establish the frame rate for the new set
of conditions. The BIOS changes made, should cause a slight loss in frame rate.
The next step, is to go to this link, and get a copy of "clockgen". The
reason I have to point you to an archived copy, is the real web page is
currently down for repairs (perhaps Franck is about to release a new version).
http://web.archive.org/web/20070502114125/http://www.cpuid.com/clockgen.php
Clockgen allows you to change the master clock, while you are in Windows. This
is different than overclocking via the BIOS. If you change the master clock
via the BIOS, and something goes wrong, clearing the CMOS would be necessary
in the odd case, to set it right again.
Clockgen has the advantage, that the messing about only exists while you are
in Windows. The BIOS doesn't have a record of the Windows clock value, and
the BIOS still has the original 200MHz value to work with. So, if Windows
crashes, as long as you can reach the reset or power button on the
front of the computer, that should be sufficient. (And I hope it won't come to
that.)
The next tool to pick up, is Orthos. Download the tool here and set it up.
Use the default "Blended" setting, and click the start button. (You may want
to trim the memory amount setting in Orthos, so it doesn't hog all the RAM. On
my machine, it uses about 700MB, which makes the starting of other programs
a bit slow. Select a smaller amount - a couple hundred MB is probably enough for
this test.)
http://sp2004.fre3.com/beta/beta2.htm
You still haven't overclocked yet. Orthos is running, and on a dual core, will
start two threads running Prime95 as a stability test of your CPU. Allow the
program to run for 10 minutes at least. The program should not throw an
error, and should run for hours without stopping. That would mean the machine
is stable before you start overclocking. If the machine is not stable, then
fix it
OK. Now unzip the clockgen program. The web page has a bit of instructions,
which may or may not be enough to get you started. Basically, what you have to
do, is tell clockgen, what kind of "PLL" you've got. Basically, that is the
clock source for your CPU. Now, Nforce4 does its own clock synthesis, which is
why you won't have to do any pesky looking at the motherboard, to see what the
clock generator chip happens to be. If you can figure out how to tell clockgen
you have an Nforce4, then the next thing that happens, is you'll see the dial
box.
http://web.archive.org/web/20070202191954/www.cpuid.com/pics/cgpllcontrol.png
The upper slider probably controls the CPU master clock input. Move the slider
a bit, and observe how the dials respond. The settings are not applied, until
you click "Apply".
The next question is, "how much do I overclock". I like to raise the control,
about 5MHz at a time. That is actually pretty important, as it allows you to
get a feel for how "close to the edge" you are getting. If you just whipped the
slider all the way to the right, Windows would crash or freeze. And in
this case, we're using "Orthos" like a canary in a mine, sniffing for
early instability.
So say you set it to "205", from "200". Click apply. What happens now, is
the clockgen program gradually raises the clock. The clock is not raised
in one big jump. I think it is raised a MHz at a time. (When I tried a big
jump on my machine, it took about 30 seconds to complete the command.)
Now, after you click the apply button. watch that copy of Orthos you are
still running. Let it run for five ot ten minutes and see if it stops with
an error. Did it stop with an error ? If it did, you've gone far enough
for now. Go back to the dial box, and turn it down again.
The idea here is, by advancing 5MHz at a time, and using Orthos to sense
the stability, you are avoiding crashing the computer. You should be
in control the whole time. (I don't recommend doing a lot of other
stressful things on the computer while doing these tests, to avoid
writing out corrupted information etc). The limit to how far to test,
is either until you are stopped by an error in Orthos, or you run out
of the slack you prepared in steps (1) and (2). In my example, that was
20%, or 240MHz. limited by the RAM.
Now, one thing that is missing from this procedure, is any description
of changing the Vcore on your processor. A real overclocker would
occasionally reboot, and drop into the BIOS, and adjust the voltage.
So far, what you're doing, is evaluating how far you can go, without
changing Vcore. If you're feeling brave, you can always drop down to
the BIOS, and crank Vcore a bit. But I've tamed my description of
the procedure, to a level sufficient to do some experiments, but
with a minimum of risk.
OK< so say you manage a 10% overclock doing this testing. You raised
the clock from 200MHz to 220MHz, lets say.
The next step is to leave Orthos running for up to four hours. That is
to determine whether the overclocked setting is really stable.
OK, now we're finished with testing the processor. Time to actually
use the result.
Dial the clock back to 200MHz. Run Crysis and get a frame rate.
Now dial the clock to the known stable value (say it was 220MHz).
Run Crysis again and get a frame rate.
If the frame rate is higher, or the game seems smoother, then
the game could be CPU limited.
In reading some more comments on various web sites, the suggestion
is, that Crysis does have a lot of stuff going on, such as physics,
that may not have been there to as great an extent in other games.
So it very well could be the processor that is partially responsible.
In the game I play for amusement, I overclocked from 2.8GHz to 3.1GHz,
and that slight an increase helped with motion blur and hesitation.
It doesn't sound like a lot (and it isn't), but if you aren't able
to meet the minumums for a game, sometimes that is all it takes to
make the game more enjoyable.
Paul
