Using the AMD Overdrive utility

[Yousuf Khan]

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

[Paul]

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

[Yousuf Khan]

On 25/11/2011 12:08 AM, Paul wrote:
> 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.

Yeah, I know, I'm really pleased that this particular processor came
down in price so nicely.

> 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.

Yup, I'm using the Stability Test inside AOD, it looks like it will
actually stop the test once it finds an error and report it. And the
error will show up within 10 minutes, so no need to run an 1 hour or
longer test.

Unfortunately, if the AOD stability test finds an error, it's already
too late, and the operating system is just moments away from instability
itself, so it will need to be rebooted. So after the reboot, I just
increase the voltage another 0.05V and usually it's stable. I've now
gotten it stable at 3.6GHz base/3.7GHz turbo. Overclocking it through
the AOD has the advantage that you can leave the BIOS settings at
completely stock, and the overclocking only takes place once the AOD
service starts up in Windows, and you can just disable the service or
just load the default settings in the service to get back to original.

I might try to overclock the Turbo speed next.

Also it looks like it's not necessary to shut off Cool'N'Quiet or C1E
states with the AOD utility, as was suggested by one of the sites I
looked up. So you can still keep a nice and cool system while idle, but
you have a higher maximum.

Yousuf Khan

[Paul]

Well, I can't say what will happen in your case, but when you use
things like CNQ, they use ACPI schema, and the multipler values
for idle and full blast, are stored in there. The question then is,
depending on how you're overclocking, does CNQ do the right thing,
or does it use the defaults for the processor ?

For example, RMClock allows you to control the response of your
processor. I think it adds an extra entry to the ACPI information,
so it can control what happens. In the Power control panel, you'll
likely see it set to something RMClock related, after RMClock is
installed.

http://cpu.rightmark.org/download.shtml

I find it's got too many controls for my liking.

http://img384.imageshack.us/img384/9124/20081108210631yy3.jpg

More fun with RMClock, here. Demo of "graphs" it draws.
That is handy for detecting CPU throttling.

http://ixbtlabs.com/articles2/cpu/intel-thermal-features-core2.html

One reason I used that tool, is it includes an "Idle Loop" task.
Back when I was using Win2K, I bought a one year subscription to
Kaspersky. Several months after installing it, I happened to make
a power measurement on my CPU (ammeter on +12V), and noticed the
consumption was too high. What was happening, is Kaspersky eliminated
the idle loop in Win2K, as some sort of protection. It meant, when the
system was idle, it wasn't using HALT instructions (an interrupt, like the
clock tick interrupt, brings the CPU out of HALT). So the processor would
wait for the next scheduled task, while sitting in some kind of loop
instead.

Ticking the box in RMClock for that, causes RMClock to run it's own
low priority task (one per core). When the system is idle, that task
runs, and the RightMark designers put HALT in their version. Basically,
it allowed overriding what the Kaspersky designers did. And it returned
my idle performance, back to where it belonged. Otherwise, I might
have not had occasion to use it that much.

I think I had some other tool, simpler in design, to fool around
with the equivalent of P-States, multipliers and VCore. And at least
that tool, I could figure out what I wanted from it. First link is
screenshot of what feature I think I was using. Second is the web site.

http://i1-win.softpedia-static.com/screenshots/CrystalCPUID_3.png

http://crystalmark.info/download/

You can scan software like that on virustotal.com if you want,
before using it. A clean reading doesn't mean anything, if
some new malware is added to a download, but it's better than
nothing. To save on uploading times, you can computer an
MD5 checksum instead, and use the "search" function on virustotal.com,
to search for previous AV scans on the file.

Have fun,
Paul

[Yousuf Khan]

On 26/11/2011 12:21 AM, Paul wrote:
> Well, I can't say what will happen in your case, but when you use
> things like CNQ, they use ACPI schema, and the multipler values
> for idle and full blast, are stored in there. The question then is,
> depending on how you're overclocking, does CNQ do the right thing,
> or does it use the defaults for the processor ?

Well, I'm using the AOD Windows utility to do the overclocking, and it
probably takes care of bypassing ACPI limits, so I do see the top speeds
changing properly. I verify it with a couple of other utilities like
Core Temp and CPU-Z.

I don't know if I were to use the BIOS to overclock it instead, would I
run into this problem? I may try overclocking through the BIOS at some
point and I'll let you know. But I somehow doubt I'd have any such
problems in BIOS either, as it is BIOS own settings that let you
overclock, and ACPI is part of BIOS, so the BIOS can probably modify
ACPI tables properly.

> One reason I used that tool, is it includes an "Idle Loop" task.
> Back when I was using Win2K, I bought a one year subscription to
> Kaspersky. Several months after installing it, I happened to make
> a power measurement on my CPU (ammeter on +12V), and noticed the
> consumption was too high. What was happening, is Kaspersky eliminated
> the idle loop in Win2K, as some sort of protection. It meant, when the
> system was idle, it wasn't using HALT instructions (an interrupt, like the
> clock tick interrupt, brings the CPU out of HALT). So the processor would
> wait for the next scheduled task, while sitting in some kind of loop
> instead.
>
> Ticking the box in RMClock for that, causes RMClock to run it's own
> low priority task (one per core). When the system is idle, that task
> runs, and the RightMark designers put HALT in their version. Basically,
> it allowed overriding what the Kaspersky designers did. And it returned
> my idle performance, back to where it belonged. Otherwise, I might
> have not had occasion to use it that much.

I don't think it's absolute necessary in my case, I'm not using
Kaspersky, and the C1E state is what takes over for the HALT
instruction. I find that when idling, the processor is throttled to
800MHz and its temperature is in the sub-30C range, usually around 26C.
The idle speed is the same even when overclocked, so it stays just as
cool as when it isn't overclocked. Highest temperatures when overclocked
has been 48C, but only 32C when at stock settings.


Yousuf Khan

[Yousuf Khan]

On 26/11/2011 1:14 AM, Yousuf Khan wrote:
> On 26/11/2011 12:21 AM, Paul wrote:
>> Well, I can't say what will happen in your case, but when you use
>> things like CNQ, they use ACPI schema, and the multipler values
>> for idle and full blast, are stored in there. The question then is,
>> depending on how you're overclocking, does CNQ do the right thing,
>> or does it use the defaults for the processor ?
>
> Well, I'm using the AOD Windows utility to do the overclocking, and it
> probably takes care of bypassing ACPI limits, so I do see the top speeds
> changing properly. I verify it with a couple of other utilities like
> Core Temp and CPU-Z.

Now, one thing I have noticed though is that even though the ACPI tables
aren't used when the AOD utility has set up the speeds initially. If you
put the computer to sleep and return from the sleep, the computer is
back to the speeds setup by BIOS. You have to rerun the AOD utility and
reload the settings again, if you want it back to overclocked settings.


Yousuf Khan

[rms]

Just to give you a data point on overclocking via the bios (which I
recommend vs some windows utility), here are my overclocked settings on a
1090T x6, on a 890GX motherboard, which it's been at for over a year now. I
used the same settings on my x4 cpu previously.

240mhz fsb
x17 multiplier = 4080mhz
x11 northbridge mult
x10 HT mult
memory x6.66 mult at 8-8-8-24 1T (corsair 1600mhz dominators)
cpu core volts 1.475v, associated chipset voltages raised also.

This is a quite high core voltage, and you must stay well below this in a
stock heatsink system. Mine is a highend Noctua NH-D14 with 2 higherspeed
fans on it, but still air-cooled.

The core & northbridge voltage (within bounds of course) is the primary
determinant of how far the cpu itself can oc, and when you raise them
temperatures go up. You *must* closely monitor temperatures when doing
stability testing, using a motherboard utility. Keep both cpu & system
temps below a max-stressed *peak* of around 55C. Your bios should have a
'Health' menu where you can force shutdown at a given temp; set this to
about 60C. When testing, if you notice temps still rising at the end of the
test run, back off the voltages a notch and set your sights lower. This is
especially important when gaming for long periods, as the videocard
temperature will rise dramatically as well as the cpu.

Overclocking SOP is to set fsb & mult as high as possible for a given core
voltage, while keeping the memory within it's stated specs by adjusting the
memory mult & fsb, then run Prime95 on all cores until it fails or gets too
hot. If it gets too hot you must decrease voltages or increase cooling. If
it fails you can bump voltages or change clock settings. Which clock
settings to change is the fun part :)

Try both, and feel free to play around with settings. You can't 'break' the
cpu with too high clock settings. You *can* break it with extreme voltage
settings or high temps, so monitor these carefully. I suggest booting off a
secondary 'trash' hard drive, not your main hard drive, until it's stable.

BTW, you should verify that Windows sees all 6 cores by entering Task
Manager, Performance tab, and check that it shows 6 cpu graphs. If it does
not, there's a simple procedure to fix that.

rms

[Yousuf Khan]

On 28/11/2011 9:12 PM, rms wrote:
> Just to give you a data point on overclocking via the bios (which I
> recommend vs some windows utility), here are my overclocked settings on
> a 1090T x6, on a 890GX motherboard, which it's been at for over a year
> now. I used the same settings on my x4 cpu previously.
>
> 240mhz fsb
> x17 multiplier = 4080mhz
> x11 northbridge mult
> x10 HT mult
> memory x6.66 mult at 8-8-8-24 1T (corsair 1600mhz dominators)
> cpu core volts 1.475v, associated chipset voltages raised also.

So you've overclocked your FSB too? Why would you do that, rather than
with multiplier alone? I assume that the 1090T is a Black Edition, I've
never heard of another kind of 1090T. Doesn't pushing the FSB affect the
stability of other components beside the CPU?

The 11x northbridge multiplier, is that the internal CPU northbridge?

> This is a quite high core voltage, and you must stay well below this in
> a stock heatsink system. Mine is a highend Noctua NH-D14 with 2
> higherspeed fans on it, but still air-cooled.

The AMD site lists a voltage range of 1.125-1.40V for the 1090T
(http://is.gd/FiNKO6). How high can you push the voltage? How high can
you push the speed using just the design-spec voltage? Do you absolutely
need to go above spec to get to 4GHz?

> The core & northbridge voltage (within bounds of course) is the primary
> determinant of how far the cpu itself can oc, and when you raise them
> temperatures go up. You *must* closely monitor temperatures when doing
> stability testing, using a motherboard utility. Keep both cpu & system
> temps below a max-stressed *peak* of around 55C. Your bios should have a
> 'Health' menu where you can force shutdown at a given temp; set this to
> about 60C. When testing, if you notice temps still rising at the end of
> the test run, back off the voltages a notch and set your sights lower.
> This is especially important when gaming for long periods, as the
> videocard temperature will rise dramatically as well as the cpu.

I've found that the Turbo Core mode stops working once I've gone over
3.6GHz. Not a huge deal, since Turbo Core is only 3.7GHz. But I've also
tried overclocking the Turbo Core to 3.8GHz, but it simply doesn't allow
Turbo beyond 3.6. This might be an internally set threshold in the CPU
that is beyond our control.

> Overclocking SOP is to set fsb & mult as high as possible for a given
> core voltage, while keeping the memory within it's stated specs by
> adjusting the memory mult & fsb, then run Prime95 on all cores until it
> fails or gets too hot. If it gets too hot you must decrease voltages or
> increase cooling. If it fails you can bump voltages or change clock
> settings. Which clock settings to change is the fun part :)

I've noticed you're pushing your RAM heavily too, is Prime95 a good
stability test for RAM too, or just CPU? The AOD's stability test is
pretty good for CPU stability testing, pushes all cores heavily, but I'm
not sure if it is a test of RAM stability though.

> BTW, you should verify that Windows sees all 6 cores by entering Task
> Manager, Performance tab, and check that it shows 6 cpu graphs. If it
> does not, there's a simple procedure to fix that.

I've got an utility called Core Temp that shows the state of all cores
continously through a Windows 7 desktop gadget.

Yousuf Khan

[willbill]

Thu, 24 Nov 2011 23:09:10 -0500, 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. ...

> 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?

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

I try wikipedia 1st, then google.

I haven't used AMD's Overdrive, or AOD as you refer to it.

Bill

[rms]

> 240mhz fsb
> x17 multiplier = 4080mhz
> x11 northbridge mult
> x10 HT mult
> memory x6.66 mult at 8-8-8-24 1T (corsair 1600mhz dominators)
> cpu core volts 1.475v, associated chipset voltages raised also.

So you've overclocked your FSB too? Why would you do that, rather than
with multiplier alone? I assume that the 1090T is a Black Edition, I've
never heard of another kind of 1090T. Doesn't pushing the FSB affect the
stability of other components beside the CPU?

FSB doesn't make as much difference as it once did, but there is a
slight (very slight) benchmark increase the higher you go.

The 11x northbridge multiplier, is that the internal CPU northbridge?

I don't pretend to know cpu architecture, but the 'northbridge' is a
separate chip from the cpu, presumably this is path between them.

> This is a quite high core voltage, and you must stay well below this in
> a stock heatsink system. Mine is a highend Noctua NH-D14 with 2
> higherspeed fans on it, but still air-cooled.

The AMD site lists a voltage range of 1.125-1.40V for the 1090T
(http://is.gd/FiNKO6). How high can you push the voltage? How high can
you push the speed using just the design-spec voltage? Do you absolutely
need to go above spec to get to 4GHz?

Unless you increase cooling with aftermarket parts, it's doubtful you'll
have to worry about exceeding the max voltage spec, you'll be hitting high
temps before that happens.

I've found that the Turbo Core mode stops working once I've gone over
3.6GHz. Not a huge deal, since Turbo Core is only 3.7GHz. But I've also
tried overclocking the Turbo Core to 3.8GHz, but it simply doesn't allow
Turbo beyond 3.6. This might be an internally set threshold in the CPU
that is beyond our control.

I disable any 'autooverclocking' option in the bios, as well as CnQ,
before ocing, making all the settings manually. You might be better off in
the end just letting it do the ocing for you though, in terms of heat and
power usage.

I've noticed you're pushing your RAM heavily too, is Prime95 a good
stability test for RAM too, or just CPU? The AOD's stability test is
pretty good for CPU stability testing, pushes all cores heavily, but I'm
not sure if it is a test of RAM stability though.

Actually my ram is running right at 1600mhz (240fsb x 6.66), the rated
spec, though I did tighten timings from 999-24 to 888-24 1T. Prime95 is a
very good ram tester imho.

> BTW, you should verify that Windows sees all 6 cores by entering Task
> Manager, Performance tab, and check that it shows 6 cpu graphs. If it
> does not, there's a simple procedure to fix that.

I've got an utility called Core Temp that shows the state of all cores
continously through a Windows 7 desktop gadget.

There's a well-known Win7 bug where it doesn't recognize extra cores
after the cpu is changed, which happened to me when I went from x4 to x6.
Do check Task Manager to verify it shows 6 core graphs.

Yousuf Khan

rms

[Yousuf Khan]

On 11/29/2011 10:21 PM, rms wrote:
> FSB doesn't make as much difference as it once did, but there is a
> slight (very slight) benchmark increase the higher you go.

With my previous processor, an X3 2.6GHz, on the exact same motherboard,
I had tried overclocking by increasing the FSB speed, it was the only
choice I had as it didn't have an unlocked multiplier. I found that once
I went past a 5% overclock (210MHz vs. 200MHz), things got unstable. Not
right away, but after several hours of use. You got a full 20% overclock
on your FSB, 240MHz!

> The 11x northbridge multiplier, is that the internal CPU northbridge?
>
> I don't pretend to know cpu architecture, but the 'northbridge' is a
> separate chip from the cpu, presumably this is path between them.

I think the concept of the separate northbridge chipset doesn't exist
anymore, as a lot of its functionality has now been taken over by the
CPU. I found out recently that AMD calls the portion of its CPU that
deals with L3, Hypertransport and memory controller the northbridge, and
it's separately powered from the rest of the CPU. Since these functions
were what used to be done by a separate northbridge chipset in the past,
AMD just usurped that terminology for inside the CPU.

> I've noticed you're pushing your RAM heavily too, is Prime95 a good
> stability test for RAM too, or just CPU? The AOD's stability test is
> pretty good for CPU stability testing, pushes all cores heavily, but I'm
> not sure if it is a test of RAM stability though.
>
> Actually my ram is running right at 1600mhz (240fsb x 6.66), the rated
> spec, though I did tighten timings from 999-24 to 888-24 1T. Prime95 is
> a very good ram tester imho.

Oh, I see, if the FSB was running at the stock 200Mhz, you'd be using an
8X multiplier instead of a 6.66X to get to 1600Mhz.

> There's a well-known Win7 bug where it doesn't recognize extra cores
> after the cpu is changed, which happened to me when I went from x4 to
> x6. Do check Task Manager to verify it shows 6 core graphs.

Oh, I see, no that wasn't a problem with me, it even installed new
processor drivers (shows "AMD Phenom II X6 1100T" instead of "X3 710"
now) after the first reboot. I guess there were just adequately
different features between the two processors to trigger a rescan.

Yousuf Khan