Undervolting the Thinkpad X120e

How exactly do you test the different p-states when using any of the stability tests use P0? Is there a tool like Rightmark CPU Clock Utility that can limit the p-state and allow intensive testing for each one like on Intel processors?

 

AMD's Vision Engine Control Center's "CPU Power" item will allow you to drag a slider to limit your computer to P0, P1, or P2 when on AC--use that to select what power state your computer will reach when you 100% stress test it.

 

Thanks! I was able to undervolt with stability as follows:

P0: 44 (1.125v)
P1: 5E (.962v)
P2: 7A (.787v)

I think I got pretty lucky with my E-350, lol

 

Wow. Yes you did!

 

Indeed--if you encounter no stability issues with those voltages, you are very lucky

On another note, I randomly BSODed today even though all three power states passed 15min of Orthos. I BSODed at--I think--somewhere around medium-high load, so I'm trying to figure out which one of my power states is the culprit. My guess is P1, so I bumped it up a notch. Testing, so we'll see...

I'm a bit surprised at the range of voltages that have worked for different people--I had expected the spread to be smaller.

 

Isn't stability testing usually run for ten hours?

 

It should definitely be more than 15 minutes. I could pass various states at 15 minutes that wouldn't run for hours. I would say 4 hours is the minimum depending on what you're using for stability validation.

 

Dang. I'm totally attempting this when I get home later. Even though I'm sitting in class with only one Chrome window up and two tabs and the fan won't even start up. Is it just me? Is there a great temperature variance for chips even when being fed the same voltages?

EDIT: Okay, ppl who have done the whole task scheduler thing, I suggest you just leave that alone. That's solid advice as mentioned above me.

Just make a new *.rw file. I called mine tester.rw
Whatever, it doesn't matter.
run "cmd.exe"
cd\
cd program files(x86)\rw-everything
In this directory, type this:
rw /Command=tester.rw

If you freeze, you'll just reboot into your good settings.

I couldn't get stable at P2: 7A (.787v)
However, I got it to accept 78 = 0.800v
All seems good so far. It's continuously streaming the French Open.

Now, how do we overclock this thing, too?
Would love my P0 to hit 2GHz and keep my P2 low like now. Is this possible?

 

Technically, if you had the time, that would be ideal.

I think even 4 hours is a little excessive, unless you want to be 100% certain you'll never hit a BSOD/lockup with your settings, but it's rather tedious if you want to test every step like that. The longest I stress tested for was 30 minutes on each of the power states (did that earlier today), and sure enough, locked up at my P1 power state. After the bump up, I didn't retest my P1, but I think it should be fine...

My advice would probably be to keep going lower until you freeze immediately upon applying the lower voltage, then back up one notch and stress test--probably the most time efficient.

 

Why wouldn't I want to be certain I won't hit a BSOD? I don't want to play a stability/voltage level guessing game. But I'm not testing every modification for four hours, of course. I poke at it with 10-15 minute tests until I get to what I think is the lowest stable voltage, and then I give that voltage point an overnight LinX run. It stretches out the process over a few days, but it doesn't really take any additional time.

I've been overclocking for a while now, and there's nothing at all fun about an unstable system. I want to be 100% certain that a lockup or BSOD isn't undervolting-related so that I can identify other problems. Plus there's the inconvenience of losing work or whatever happens when a system locks up in the middle of something. It's just not worth the hassle not to test.

It's not that bad to be pretty sure something is stable and then give voltage a bump if you hit a problem, but what makes it tricky when you're undervolting different power states is that it's tough to be sure which state (or which combination of states) is the problem.

 

I'm happy to report 0.800v successful (limited to P2 only)
4 hours of streaming ESPN3 at highest quality, no issues.

 

Sugarkang could I possibly have a copy and paste of the file that is working for you so I can insert it into mine. Cheers mate.

 

Okay, anyone new to this thread, please see MidnightSun's first post. Follow those instructions. Use those settings first.

Here's a little screenshot you might enjoy.



It's the Ortho test, but oh yeah...
with two videos at the same time!
I ran this torture test on all of these settings:

P0 = 0.8v
P1 = I forgot
P2 = 1.15v

Your results will probably be different. AGAIN. Consult MidnightSun's first post and do that before you try mine.

 

Ah, I forgot that Orthos is basically multithreaded Prime95. I've always considered 8 hours to be the stability mark for Prime95. I've had a number of machines that would chug along happily for more than four hours, only to be unstable running other tasks.

 

Someone gonna tell me if we can overclock the high end while keeping the undervolt settings?

Heck, I'd even be willing to go higher voltage on P2 if we can get 2Ghz. Come on, somebody knows the answer. People used to do this on desktops years ago...

 

I poked at it a bit. You can underclock the highest p-state, and you can probably overclock the intermediate states as high as the max state, but you can't push the max p-state multiplier any higher - it just wraps around to the bottom.

 

sugarkang unfortunately mine kept freezing up with your undervolt settings, you have a great machine there! I've reverted mine back to the original on the first page (for now). Would sure love mine to go as low as yours has!

 

Just noticed my cores are running at 807.2Mhz is this normal?




http://imageshack.us/photo/my-images/713/capturefkt.png/

 

Yes, that's the whole point of power state P2--lower voltage/power consumption, and lower clocks when you don't need it or when you are on battery and need to save power.

 

Yeah, don't even sweat it.

1.037v = default voltage
0.825v = your voltage (about a 20% drop)
0.800v = my voltage (an extra 2%?)

I may get bragging rights, but I noticed may 1 degree temperature drop?
You got most of the benefits.

 

I did the same with my Ideapad S205, using this awesome thread.

These are my values (absolutely stable with Orthos running for 3 hours):

Code:

 >cpu 1
>wrmsr 0xc0010064 0x80000126 0x00004210
>wrmsr 0xc0010065 0x8000011A 0x00005C12
>wrmsr 0xc0010066 0x8000028C 0x00007630
>cpu 2
>wrmsr 0xc0010064 0x80000126 0x00004210
>wrmsr 0xc0010065 0x8000011A 0x00005C12
>wrmsr 0xc0010066 0x8000028C 0x00007630
>RwExit 

which means 1.137 V for P0, 0.975 V for P1 und 0.812 V for P2. I guess i got an undervolting-friendly E-350

BTW, i tested those values for each voltage using Lenovo's "Energy Management" tool. It has four profiles:
"Energy Star": Switched between P2 (~800 MHz), P1 (~1300 MHz) and P0 (~1600 MHz). Useless in this case.
"Performance": Permanently on 1.6 GHz. Useful for testing the first value (P0), 0x00004410.
"Balanced": Switches between 800 and 1300 MHz. Useful for testing the second value (P1), 0x00005C12.
"Powersaving": Permanently on 800 MHz. Useful for testing the third value (P2), 0x00007630.

I didn't hear anything when Orthos was running. When i moved the ventilation slit to my ear, i hardly heard the very quiet, slightly spinning fan.

Unfortunately, TPFanControl doesn't work for me.

 

Interestingly I initially installed a clean Windows Home Premium 64 and my fan was intermittently on 60% ish of the time. I undervolted with standard settings on the first page and used custom and standard settings with TP Fan Control.

Yesterday I installed a clean Windows Ultimate 64, installed the same drivers, undervolted with standard settings and kept standard settings with TP Fan Control.

My fan is less than 20% on and I can't seem to work out why. It definitely does not come on as much as before. I'm happy nonetheless.

 

Is there perhaps a way to change the default bus speed then, like in the good old days? I'm not familiar with how these CPUs derive their clock speed...

 

It might be possible with SetFSB if you know the PLL. That's usually how it's done.

 

Thanks for the tip! For some reason TPFanControl doesn't start on startup for me no matter what I tried.

Also overclock would be nice too.

 

Hello everybody. First of all, thanks to MidnightSun for this thread, very useful and easy to understand.

I own a Sony Vaio VPC-YB1S1E that also has the new AMD E-350 APU. I did everything like you said in your first post and this seems to work for the Vaio too! I got 0.825V in idle and 1.2V with load (video playback) - so everything should be working as it should right?

Now i would like to test how low i can get wih the voltage, i cant really damage anything by undervolting, can i? The only thing that could happen is that the device gets unstable and i start getting blue screens/crashes right?

By the way, how much longer battery life can i expect from undervolting?

Thank you.

 

You're very welcome

I actually took many of the details from the SPCR thread on the dm1z, so I would expect that much of the settings carry over to other E-350 "netbooks" such as the Sony Vaio YB, Dell Inspiron M102z, etc. Glad it's working for you--yes, those are the "standard" undervolt settings I have in my guide. With the tolerances of each individual chip, though, your max undervolt can be either higher or lower than that value.

No, the most you'll get is lockups and general crashes. Follow the instructions in my second post to lower the voltage for each power state incrementally and test using Orthos, until you get an immediate lockup. When you get that, bump up the voltage and do an intensive test to see if the next highest voltage is stable--if you get any instability, you'll need to increase one notch again and retest.

As for battery life, that depends on how far you can undervolt, and also on your usage. Personally, I haven't noticed a radical difference in battery life; however, the fan turns on far less often and the system runs cooler, which was the main thing I was after.

Good luck.

 

I will try how low i can get with this laptop and post my results in here later. With the configuration from the first post, i already got 3-4 degrees cooler cpu temperature and the fan runs at less rpm

However i am still trying to find out how i can control the fan via software. If i find something i will post it here.

(btw: you are all very lucky to have the x120e, here in germany they dont sell that nice piece of hardware! so i had to get the sony)

 

You can buy a TP x120e on eBay, there are some US/Canada sellers who ship to Germany. For sure you'll have to pay the duty...
I also found a German seller.

 

would it, in theory, be possible to set the p state clocks by the same method (write the right values to the right msr register) and thus over-/underclock?

 

You can do that to move the multiplier down and underclock, but it's not possible to overclock.

 

had a quick look at the amd documentation and it would seem on might be able to change the NB's PLL frequency that presumably controls core frequencies by changing MainPllOpFreqId located at "D18F3xD4 Clock Power/Timing Control 0"
so probably something along the lines of

followed by a p-state change or something might do the trick, where 0x40 is the minimum (I'd assume 0x50 to be the default but i don't know). Maybe one also has to change some clock divisors, no idea and i can't play with it as i don't have a e-xxx board/machine yet but as gigabyte and asus offer overclocking on their mini-itx boards and the northbridge is fully part of the APU i don't see why it shouldn't be somehow possible to under-/overclock.

 

Hi everyone,

I have modified an older program (PhenomMsrTweaker by Martin Kinkelin), which was used to set VID and FID values for the different P-States on Phenom systems. It now works on different Ontario laptops. If you want to give it a shot, here are the links:
Win32: http://dl.dropbox.com/u/4722708/PhenomMsrTweaker x86 Setup Ontario V1.msi
Win64: http://dl.dropbox.com/u/4722708/PhenomMsrTweaker x64 Setup Ontario V1.msi

The biggest part of the development was definitely done by Martin. Some more explanations can be found here:
PhenomMsrTweaker: Cool and Quiet customized - XtremeSystems Forums

Features:

With this tool you can simply adjust VID and DIV (a.k.a. FID on Phenom) for each P-State, while you can also overclock the Ontario APU (at your own risk). Once you are done and you know the best settings, click on Service.. + Update and make the settings permanent, which then always get applied on Win startup through a Windows service just in the background.

Regards,
witteks

Edit: Updated links

 

witteks,

I gave the program a shot.. It changes the multiplier, but i see no difference in the performance that much.

Here is the stock performance:


Uploaded with ImageShack.us

Here is the 2.1GHz Performance:


Uploaded with ImageShack.us

I gotta do more benchmarks, but something feels fishy about this.

 

Maybe CPU-Z is incorrectly reporting the speed? I've seen reviews of overclockable E-350 boards where the reviewers had difficulty getting much above 1.7GHz or so. 2.1GHz seems like an unrealistic overclock for the CPU.

Try running a benchmark like SuperPi to test the difference - that should be more accurate.

 

Tested this on a S205
Undervoltung and UNDERclocking defently works, Overclocking is not working, the shown values are wrong.

 

You are right, that overclocking is obviously not working. CPU-Z is reporting a wrong Mult (and I was relying on it ), because in the background the bus speed seems to get throttled by the CPU itself. So, while having a Div of 1, which should theoretically yield 3.2GHz, the bus speed gets changed to 50MHz. At least, this is what other guys reported from using the AIDA tool.

Actually, I still don't know, why this is happening, since I couldn't find a hint of that behavior in the AMD specs. However, I will try to find the register, which AIDA reads to get 50MHz and may be there is an opportunity to play with that.

Regards,
witteks

 

Would be nice!

 

Btw, thanks hellsy for giving it a shot

 

I'm fairly sure the E-350 has a locked multi, as in you can not raise the multi, just lower it. That limits any overclocking to bumping up the base clock (FSB). If anyone can find a tool to change that, overclocking shouldn't be too hard. Does AMD OverDrive work on Zacate? (Or any laptop for that matter, I've only used it on my desktop). It should allow for messing around with the settings for the GPU as well.

 

No problem witteks.. We are all trying to tweak this little machine. It felt too good to be true for it to work at 2.5GHz at 1.3v. 900MHz would show a great performance increase..

 

The point is, that the E-350 doesn't have a locked Multi. The Multi is fixed for all P-States and you can change the divider, which is what I did. That functionality is different from Phenom.

Here is the equation:

FSB * 32 / Divider = Core Frequency (E-350)

FSB * 40 / Divider = Core Frequency (C-50)

There seems to be some software, which makes sure, that the Core Frequency doesn't exceed 1.6GHz (E-350) Core Frequency by throttling the FSB. I couldn't find something in the spec so far, who and how this is done.

-witteks

 

"You can’t change the processor clock frequency multiplier for AMD E-350" -XBit

Maybe the behavior of the base clock halving when you have div at 1 is their mechanism for "locking" the multi. Does this behavior happen when raising the div? (If more div, base clock goes up to compensate). If so, would this influence memory speed (since I'm pretty sure RAM is based off of base clock).

 

No, it doesn't get raised. It stays at 100MHz.

The more I think of it, the more I doubt the lowering of the FSB actually, which can be seen in AIDA. It might be the case, that the effective Core frequency stays at max 1.6GHz, since benchmark values don't change much, which is what I expect, if the FSB would be halved. Maybe the truth is somewhere in between the reported data from CPU-Z and AIDA.

But there is no overclocking possible right now.

Regards,
witteks

 

As any NON-BE CPU, and Zacate being a low end APU, i think the upper multipliers are pretty much locked down.. I think the program we tried basically modifies something at the software level making the software report different then the original value. I am not a programmer, but if someone figures out what clock generator Lenovo used for this motherboard, and SetFSB team takes a look at that and adds the support, i am pretty sure we can go a long way .

 

Anyone tried SetFSB with the X100e PLL setting?

 

Good point. but I'm not certain, if Lenovo is using the same clock generator for 100MHz as for the X100e, which runs 200MHz.
Worth a shot though.

Regards,
witteks

 

I'm having no luck getting this mod to work. CPU-Z reports stock voltage for all power states, though I've double-checked all procedures and task scheduler indicates the task is running. Switching between "balanced" power profile and my custom Lenovo PM profile likewise has no effect.

I was not able to create the task as a standard user and had to login as administrator to create the task. I have to run task scheduler as administrator to even see the task in the list...it doesn't show up otherwise. Likewise, I have to run CPU-Z as administrator otherwise it throws an error (0x5) and won't report CPU speed and voltage. However if I'm logged in as administrator CPU-Z still shows stock voltages.

 

In your custom Lenovo Power Manager profile, is the CPU set to Adaptive speed? If not, change it to that and see if the lower voltages are applied within a few minutes after boot.

 

Hi tychoseven,

You can give the tool from post #83 a chance. It takes care of the voltages and divider, though it can not overclock

regards,
witteks