View Single Post
  #1  
Old November 25th 11, 06:08 AM posted to alt.comp.hardware.overclocking.amd
Paul
external usenet poster
 
Posts: 13,364
Default Using the AMD Overdrive utility

Yousuf Khan wrote:
I just picked up a Phenom II X6 1100T Black Edition. Now that this
processor family has become the previous generation processors, I was
able to pick this one up fairly cheaply. What used to be a $300+ chip, I
got now for less than $200! It drops in and replaces my previous Phenom
II X3 710 processor. The older Phm II was a 2.6GHz processor, while this
one is Turbo-cored a 3.3/3.7GHz processor, not to mention the newer one
has twice the cores. So I think it was a nice upgrade all on its own,
without overclocking.

I don't normally overclock processors anymore, but seeing as I have a
Black Edition, I feel I'm committing a form of blasphemy not to even try.

Last time I overclocked a processor was a 486, and so I'm a little rusty
with the modern procedures. I'd like to overclock within the limitations
of my *stock* heatsink, so I'm not expecting extreme overclocks, merely
respectable. But the meagre attempts I've done so far seem a little
pathetic. In the BIOS, I was only able to achieve a fully stable
overclock by raising the clock multiplier from 16.5 to 17 (i.e. 3.3GHz
to 3.4GHz, yee-haw). Going to higher multiples results in the machine
rebooting itself after a few hours. I've tried it upto 3.7GHz. Maybe I
needed to change some voltages too, but I didn't touch those.

So then I decided to put everything back to stock in the BIOS, and
decided to try the AMD Overdrive (AOD) utility. It has a built-in
stability test, which was the primary reason I downloaded it while I was
doing the BIOS overclocks. But then I realized that it has its own
overclocking facility, and it's a lot quicker to make changes here than
rebooting and going to BIOS all of the time. Can somebody give me tips
on how to proceed with the AOD step-by-step? How important is it to
change the voltages as I overclock, and by how much should I raise the
voltages if any? Also I found that after running AOD's stability test
for an hour, AOD seems to disappear after I come back. So is that normal
behaviour for AOD to disappear after running the stability test, or
should it just sit idly and let you know that it's finished the
stability test?

AOD suggests that I turn off the Turbo feature of the processor, the
Cool'N'Quiet, and turn off the C1E sleep state. What is the reason for
turning these off features? Do they impact your ability to overclock?
What is the C1E state anyways, it's the first I've heard of it?

Yousuf Khan


You don't need anything too fancy.

In your case, you're "Driving a Cadillac". You have multiplier
control, so issues about accidentally overclocked RAM are a non-issue.
You'll need to use some VCore, to get to use more steps.

Basically, the process as I see it, is to "draw a speed versus VCore voltage graph".
Raise the CPU speed in tiny steps. Boot to your OS (the tiny steps are
important to not outright crashing the OS immediately).

I run Prime95 for testing. I gradually inch up the speed, until I
see an error in Prime95 in five to ten minutes. Then, I have a "data point".
I bump the voltage a bit, and retest. If it's stable, move up in frequency
again, until it becomes slightly unstable and throws another Prime95 error.

On the processor I tried this on years ago, eventually I "hit a wall".
At a certain frequency, I got my Prime95 error. I increased VCore
and it wouldn't go away. I increased VCore again, and still no progress.
I went to a step below the "unsafe" voltage for a final check, and the
graph now had a "wall" in it.

There is no point in being right up against that wall.

Some processors, don't have a wall, and instead get uncomfortably
hot before hitting a limit. The Intel D805 would go to 4GHz and
draw 200 watts. And some motherboards got so hot by doing that,
that the VCore components would melt foam rubber used to support the
test motherboard. You don't want to go that far.

Since I'd hit the wall with my processor, then I could back off the
frequency, and set VCore to a step more than the value in the graph.
That's my "margin". Next, run Prime95 for four to eight hours. If no
errors, you're done. Maybe you need another voltage step, if it's still
a bit wobbly. You can throw in a simultaneous 3D looping benchmark as
well as Prime95, for further stress of the computer.

Now, with the multiplier approach alone, you've lost granularity.
You could mix both clock adjustments and multiplier, but if you
do so, then you have to sit down and do the math for the RAM settings.
A clock change affects both CPU and RAM. The RAM readout in the BIOS
will likely not be reading "true speed". You have to compute the true
value yourself, while sitting in front of the computer. If you're offered,
say DDR3-1066 and DDR3-800, these represent different ratios between
CPU input clock, and final DRAM clock. If bumping up the CPU clock, pushed the
RAM above DDR3-1066 true value, you use the DDR3-800 BIOS setting, to
select a different ratio, and then the new value is the ratio of clocks,
times the reduced ratio value selected. In that example, you can get
a lot of testing out of a 1066 to 800 change, without violating the limits
of the RAM. (You may also need to work out appropriate values for Tras,
Trcd and so on.)

If the input clock to the CPU was 200MHz, you might proceed in 5MHz steps,
while watching for Prime95 errors. If you see no errors in Prime95 in ten
minutes, you bump the clock again. Since you have multiplier, you could scan
with clock, until you get to the value you could have had with one multiplier
step, then go back to 200Mhz clock, while using +1 on multiplier. The
clock adjustment, gives you finer granularity between multiplier steps.

The reason for using tiny steps between experiments, is so you
don't crash the computer, and remain in control the whole time.

I've had the computer so unstable, Prime95 errors out virtually immediately.
And you're working in too large a step size, if that happens.

If you have a Windows overclocker program, that saves the reboot step. The
online overclocker application I used, would make slow clock changes (slewing),
and for the program to take a giant step, it cranks the clock gradually
over a 30 second period. That isn't absolutely necessary, but it's
fun to watch.

Anyway, in my view, that's the thing to do. Watch TV with one eyeball, and
every ten minutes, check to see if it's time to tweak the computer. It'll
take a while to draw the graph of frequency versus VCore, but once you have
the graph in hand, you can select close to the final values, and run a
longer term stability test overnight.

The reason for disabling Turbo, CNQ and C1E, is so you're in absolute
control of things. Using CPU-Z, you can check the core clock speed while
you're working, and verify the value. Various power saving options,
could tempt the system to enter a different P-state, many times a
second, and you might not be sure of your result. On the computer
I'm typing on, I have a whole bunch of C options that have to be
disabled, before the CPU clock stands still at the canonical value.
I have a different brand of motherboard, that doesn't take nearly
as much to "tame" and run at a fixed rate.

Just before you "turn on Prime95", you wouldn't want the motherboard
making any sudden leaps in frequency while idle. You want as
steady a frequency as possible, to better help you prepare your
SHMOO plot.

http://en.wikipedia.org/wiki/Shmoo_plot

(Example SHMOO...)
http://www.realworldtech.com/page.cfm?NewsID=354

The C states are mentioned here, but some motherboards go all the
way to C6, so this article isn't complete. You may find more info
on the C states, on another site.

http://en.wikipedia.org/wiki/C1E#Processor_states

Paul