Awesome Llano Overclocking Tool

[NdMk2o1o]

Well wait no longer!

As i feared "overclocking" does bugger all :shadedshu

sandra cpu, wprime, super pi, im not bothering with anything else until NdMk2o1o comes back with some numbers that suggest it works then we can figure out if there is an undocumented step or something.

Sorry completely forgot about this, I did run Super Pi, no need to run anything else as there was zero difference running at stock compared to supposed OC did have a placebo effect though, does that count?

 

[Mathragh]

Sorry completely forgot about this, I did run Super Pi, no need to run anything else as there was zero difference running at stock compared to supposed OC did have a placebo effect though, does that count?

Awww, bummer.

Ah well, another mistery solved i guess.

 

[Mussels]

ah well, at least we know it works on llano.

 

[HTC]

ah well, at least we know it works on llano.

According to post #83 (check posts #86 and #88 as well) it doesn't work on all Llanos

Since then, i tried not messing with the multi and only lower the volts (giving much higher then what's seen in post #83) and it still crashes

Either it's a bug in the program or something in my particular hardware that prevents the program to function properly

 

[Mussels]

According to post #83 (check posts #86 and #88 as well) it doesn't work on all Llanos

Since then, i tried not messing with the multi and only lower the volts (giving much higher then what's seen in post #83) and it still crashes

Either it's a bug in the program or something in my particular hardware that prevents the program to function properly

mobile or desktop? seems like its the mobile ones that are teh best.

 

[HTC]

mobile or desktop? seems like its the mobile ones that are teh best.

Desktop.

 

[Mathragh]

Yo_Wattup also has a desktop chip, and he had some unbelievable results, apart from the fact that his RAM sucked, lol.

So i wouldnt count out desktop chips, even more, since they're actually exactly the same sillicon.
The only difference between desktop and laptop chips is how their P-states have been set(and perhaps slightly different leak-currents, favoring lower leak currents for laptop chips).

 

[MadduckUK]

Sorry completely forgot about this, I did run Super Pi, no need to run anything else as there was zero difference running at stock compared to supposed OC did have a placebo effect though, does that count?

time for some undervolting then, if i cant have speed over stock i'll bloody well have battery life over it

 

[Mathragh]

time for some undervolting then, if i cant have speed over stock i'll bloody well have battery life over it

Haha, thats the spirit!

Awaiting some results!

 

[Yo_Wattup]

there is no ram adjustment of any sort, Yo_Wattup is most probably well aware of this. is it some sort of forum in-joke?

I am not well aware of that, why would there not be an option in bios to adjust your memory? I was just trying to help.. don't get your panties in a bunch..

 

[MadduckUK]

I am not well aware of that, why would there not be an option in bios to adjust your memory? I was just trying to help.. don't get your panties in a bunch..

you sound SEETHING

 

[HTC]

Managed to get it to work but @ much higher voltages then before.

Default voltages:

Current voltages:

Some testing using IBT and 4 GB RAM:

Not so sure is stable because, while i was trying to test some voltages, it crashed right after successfully passing IBT, which leads me to believe this test doesn't work so well for stability proofing, @ least not for APUs.

 

[Yo_Wattup]

you sound SEETHING

*facepalm*

Ok.

For the record are you a female?

 

[Mathragh]

Managed to get it to work but @ much higher voltages then before.

Not so sure is stable because, while i was trying to test some voltages, it crashed right after successfully passing IBT, which leads me to believe this test doesn't work so well for stability proofing, @ least not for APUs.

Those voltages seem way more plausible tbh, Thats about as much of a voltage drop as mine does.

Also, for real stability testing, you should also test the turbo seperately from full load, as turbo usually doesnt kick in when all 4 threads are at full load. Also, testing your APU with both the CPU, and GPU part fully loaded will be the best way to test for stability at full load, since only then the temperature will reach the highest they can go.

Another point; since turbo doesnt really kick in when the chip is under the most load anyway, you can be safe with higher voltages, and thus a higher frequency, without it costing you alot of power and heat. So I recon you can get alot more out of that turbo state!
(you can start by just leaving the turbo voltage at what it was, and testing for the max stable frequency at that voltage)

Keep it up!

 

[HTC]

Those voltages seem way more plausible tbh, Thats about as much of a voltage drop as mine does.

Also, for real stability testing, you should also test the turbo seperately from full load, as turbo usually doesnt kick in when all 4 threads are at full load. Also, testing your APU with both the CPU, and GPU part fully loaded will be the best way to test for stability at full load, since only then the temperature will reach the highest they can go.

Another point; since turbo doesnt really kick in when the chip is under the most load anyway, you can be safe with higher voltages, and thus a higher frequency, without it costing you alot of power and heat. So I recon you can get alot more out of that turbo state!
(you can start by just leaving the turbo voltage at what it was, and testing for the max stable frequency at that voltage)

Keep it up!

Had to increase one voltage increment in all PStates: it was crashing and i didn't know which PState was the culprit but, for now, seems more stable.

Got all PStates to match P7 and loaded IBT: once it passed, got P1-P6 to P6 values and loaded IBT again. Did this all the way to P0. Does this not test the turbo properly?

As for the allot more from the turbo part, OCing was never my intention: cutting power was.

 

[Mathragh]

Had to increase one voltage increment in all PStates: it was crashing and i didn't know which PState was the culprit but, for now, seems more stable.

Got all PStates to match P7 and loaded IBT: once it passed, got P1-P6 to P6 values and loaded IBT again. Did this all the way to P0. Does this not test the turbo properly?

As for the allot more from the turbo part, OCing was never my intention: cutting power was.

How you test your turbo state is the most fail-safe, but it is also overdoing it a bit. You are testing the turbo speed on all 4 cores, while in real life in these APU's, it will never turbo all 4 cores if they are loaded fully(meaning the temperature will be a bit lower, although i guess this is more important in laptops than it is for desktops).

Also, a higher Turbo state may actually lower power consumption a bit, as it will mean(atleast in some cases) that the processor may be going to a lower powerstate quicker aswell, and thus will stay more of its time in a idle, low power state. Also, if you want to lower your power useage, you can always lower your lowest P state multiplier to something like 5 instead of 8. This wil probably allow you to drop the voltage to somewhere around the 0,65 volts.

 

[HTC]

How you test your turbo state is the most fail-safe, but it is also overdoing it a bit. You are testing the turbo speed on all 4 cores, while in real life in these APU's, it will never turbo all 4 cores if they are loaded fully(meaning the temperature will be a bit lower, although i guess this is more important in laptops than it is for desktops).

Also, a higher Turbo state may actually lower power consumption a bit, as it will mean(atleast in some cases) that the processor may be going to a lower powerstate quicker aswell, and thus will stay more of its time in a idle, low power state. Also, if you want to lower your power useage, you can always lower your lowest P state multiplier to something like 5 instead of 8. This wil probably allow you to drop the voltage to somewhere around the 0,65 volts.

That's what i did @ first (check post #83): didn't work so well

Dunno what's the default for laptops but for my APU, all of the PStates have the 4 cores @ the same multi (check pic below): doesn't this mean that the turbo, when it kicks in, does so with all 4 cores?

 

[Mathragh]

That's what i did @ first (check post #83): didn't work so well :

Well, I guess i can only say what I did to get to the P-state and voltage that I ended up:

First of all i just set the lowest P state mult to '5'.
Then i get into the advanced power management section of windows, and set the maximum processor frequency of my current power scheme to something like '5%', so the APU wouldnt go higher than its lowest P-state.
After that i just started lowering the voltage, and noting the voltage I was on, untill i got a bluescreen.
Then i booted back up, and set the voltages 2 steps higher than where I bluescreened.

Its been stable for over a month now this way

Dunno what's the default for laptops but for my APU, all of the PStates have the 4 cores @ the same multi (check pic below): doesn't this mean that the turbo, when it kicks in, does so with all 4 cores?

http://img.techpowerup.org/120213/2012-02-11_130818.png

Nope, that only means that the turbo will turbo all the cores up to the same frequency, (instead of like, one core turbo-ing to 2,9GHz, while another goes only up to 2,8GHz) Llano's can alter the P-state for all cores seperately I think.

 

[Mussels]

in the program i use, turbo has its own P state and a checkbox to disable it. the reason it has it visible on all four cores is because any two of the four can run at turbo - and they swap in real time.

its not like core 0 + 1 turbo and the threads are moved to it, the cores the active threads are on have turbo go to them - and if those threads move, so does the turbo.

 

[Steevo]

Thats really weird.

The only thing I can suggest then is retry the lowest voltage for every P-state.

I'm not sure whats causing the instability, it could be heat(although I doubt it), it could be a bios issue, or even a motherboard issue(Voltage regulators getting weird at low volts(I dont have a clue)).



Also remember that NdMk2o1o has a newer Brazos, the E450(clocked at 1,65GHz instead of 1,60GHz). Perhaps this explains why he can overclock, while others have failed before(with a E350)

Mine did the same, it was caused by the miniscule time difference in the request for more power and the actual delivery at the next speed/power state, bumping up the voltage cured it. I think it is also related to a weaker core that causes it.

I ran OCCT for a hour each on my lowered voltage and tweaked speeds, but still got the occasional crash, and just bumping the voltage back up a few notches from where it was stable provided the fix.




Now we need to find out how to overclock the graphics core.

 

[Mussels]

Mine did the same, it was caused by the miniscule time difference in the request for more power and the actual delivery at the next speed/power state, bumping up the voltage cured it. I think it is also related to a weaker core that causes it.

I ran OCCT for a hour each on my lowered voltage and tweaked speeds, but still got the occasional crash, and just bumping the voltage back up a few notches from where it was stable provided the fix.




Now we need to find out how to overclock the graphics core.

i had to do the same thing actually, with the higher than stable voltages.

and yeah i wanna OC my APU so bad

 

[Modestweaker]

I want to thank Mathragh for starting this discussion. This is the exact kind of information I was looking for. I have a Lenovo with the A63400M and I was wondering if it could live up to the AMD expectation for overclocking. I must report that I am pleased.

I ran SuperPi with stock settings getting the following numbers:

Then I bumped up my P0 profile to 3GHz and the P1 set to 2GHz. This seemed like a generous overclock without pushing the limits all day. Instant win:

Did I do my math wrong is this a %43 performance increase?

 

[Mathragh]

Very nice results!!

great overclock man^^

Be sure to also give undervolting a try, most of us in this thread have been quite succesfull in lowering the power use at the same time as overclocking.

And its always nice to hear someone found my post usefull thanks!

 

[Modestweaker]

Thanks,
I have turned down my idle VID until I got bluescreen. Right now I am idling at 600MHz with a VID of 0.775v.

I found through observation that my computer spends most of it's time either at P7 or P1 so these are the settings that I gave more aggressive voltage underclocks. My P1 is rocking solid at 2GHz using only 1.025V. With a heavy load on it the temperature stays about 59C after the fans pick up.

 

[m&m's]

May someone run 3DMark 11 in Performance preset? With before and after overclocking? I would like the see the Detailed scores,
(3DMark Score, Graphics Score, Physics Score, Combined Score, GraphicsTest1, GraphicsTest2, GraphicsTest3, GraphicsTest4, PhysicsTest and CombinedTest). I would to see the gain in 3D that you got by overclocking your Llano!