[To all newer AMD Turion owners]Strange power handling on hp dv5

Well this is pretty strange, but sometimes happens that under different loads the processor behaves differently.

Anyway, my idea is that smaller machines like tx2 should not host 35W processors like ZM turions. HP technician once told me that notebooks aren't designed for high and sustained loads... it's a nonsense, but hey, they told me that consumer notebooks should not run games or cpu-intensive tasks Maybe next time they have to tell me why they put faster cpus in their machines...

However, if you don't mind the small performance drop, you may underclock and undervolt your processor at highest pstate. Depending on the game, it should have neglible to null impact in performances.
Maybe just dropping to 2200-2000 Mhz will give you some less heat, and maybe you will be able to downvolt more.

 

LOL.
Justification for flaw in design...
Why don't they just sell calculators...
You can seriously consider a suit against them.

 

Lol, yes they're funny as hell. THey told me the same

Yup!

 

i try to underclock my processor,and i surprisely notice that if i change pstate 0 below 1.1v,any frequency including 2.4GHz will lead to crush

dose 2.2GHz really mean less heat and more stable than 2.4GHz under the same voltage?

 

i was totally crushed.
i find that if i use firefox,the laptop will crush when i set the voltage at 1.1v,but 1.125v will be fine.......any suggestion?

 

Unfortunately there are some great mysteries with these processors.

For example, if I give 0.850v at 1050 Mhz in pstate 1 everything goes fine.
If I give 0.850v at 1100 Mhz in pstate 0, everything goes to hell.
There's just a 50 Mhz difference at the same voltage, but different pstate (and different DID, btw). Honestly I don't understand why the processor crashes, probably this is due to the fact that DID=0 (higher frequencies) requires more voltage than DID=1 (lower frequencies) and it is not directly related to frequency itself.

Also it is very strange that these processors have very tight undervolting abilities, for example my zm-80 has 1.100v as default, and I can't make it stable pulling it below 1.0875v.

Anyway, did you apply the pstate6/7 patch I talk about in the first page?

 

Yes,that is the only thing i can change to my cpu now ,and i make a .bat to make sure it will run correctly when computer startup.

Further more,see this

Is the HTC Limit temperature and HTC Hysteresis temperature too high?
And another question:What's the difference between the two temperature above?

 

When the processor goes above HTC Limit temperature, it reduces its frequency to 800 Mhz. This means that the processor is in a overheated state.
When the processor is in 800 Mhz state, it has to go below the HTC Hysteresis temperature to restore its original frequency.

This is specified in the datasheets. Actually I made some tests, and noticed that HTC Hysteresis temperature in not taken in account at all. If the temperature is above HTC limit, the processor is capped at 800 Mhz. If temperature is below HTC limit, the processor just runs at full speed.

You can control HTC features with the right flags using TurionPowerControl. I also noticed that HTC (that means Hardware Thermal Control) and pstate 1 undervolting causes system freezes when the processor goes overheating (bad voltage/frequency switch timings probably).

Don't know what to say actually. Maybe you can paste the output of TurionPowerControl with -l switch to let me see if there's something that could be interesting.

 

here it is.Hope you can find something intresting

Do you notice this

What about this suppose to mean?

 

IMHO, the ZM-series USES 3 pstates, o, 1 and 2, for a count of 3 power planes.
Maybe it's what it's meant.
Shot in the dark!

 

Unfortunately I see nothing of real interest, just the common register values found on ZM processors

It means that your processor has pstates 0, 1 and 2 enabled and activated, so your processor is able to switch between three different operating frequencies.

BTW you may try to downvolt pstates 1 and 2 too, they also can give nice reduction in power consumption.
For example, I can downvolt psate 1 from 0.9500v to 0.775v with full stability.

 

Well,i'll try it.
PS: Does this help? http://support.microsoft.com/?scid=kb;en-us;974090&x=3&y=4

 

Sure it can be useful. I don't know if it is included in the automatic updates lists, but surely it could worth installing it.

 

I clicked on that link, and although there is info on that hotfix, I didn't manage to find the download for it.
It's a 64 bits hotfix for AMD CPUs
It says Vista SP2 64 bits is required in order to install that hotfix.
I have all the available updates installed, so I assume I must have that hotfix installed/covered.

 

Now that I just received my Win7 upgrade kit, I installed 7 and everything seems fine.

I now tried to get K10stat to run at startup like in Vista, using Weinter's guide.
I got it to work fine in Vista. Now, the same commmands don't work in 7.

I can still run K10 stat manually at startup, but I want "automated"!

Any ideas?

 

If you followed the guide then it might be running as a system task, which means the icon won't show in the tray. Try opening task manager and showing all processes to see if it is listed.

 

Put shortcut into startup folder with in Target "C:\Program Files\etc...\K10STAT.exe" -lp:1 -StayOnTray -nw

 

EDIT: KK I see what you're doing now. I'll try this instead.

Do I still put the k10stat.dat file in the system32 folder?

Alright Thanks all I got it going.

This is what I did:
After doing this :

I deleted the k10stat.dat file, re-created another one with my settings. Rebooted the comp. Then put the .dat file in system32. Reboot. Restart and then it worked...
Jeez...seems like I tried this several times...

Anyway, thanks.

For the pstates, I got 2200mhz p0: 1.025V not stable. 1.05V is ok.
1100mhz p1: 0.8V
550 mhz p2: 0.75V
CPU ZM-82 2.2GHz

 

I'm not sure that anybody will be interested, but I started gui for your program. I had a working very simple version 2 hours ago, but right now I've got segfault and it's a little late in here, so I can't repair it right now.

It's Qt based, so I'm almost sure it will be working on Windows too.

I've got 2 questions... can I do any change to your code? can I license it on GPL v3? Of course you would be mention as the main author of it

 

Well sure I'm interested
Do all modifications you want without problems. About licenses, think I want this program will be always open and freely modificable by others.

 

Ah, i'm glad someone decided to do what i started and never finished. And my GUI was to be Windows-only so one more point for you.

 

I'm glad to hear that...

So the GPL is really good option, it prevent from closing software, it's always open and it can be freely modificable by anyone, and those modifications have to be open to. You can read it here, it's not so long.

What did you completed? I've got for now changing the VID, FID, DID and that's all, so maybe you done more than me

 

@blackshard83

I love your program, but i'd like to see TurionPowerControl working with Turion 64 X2 "Tyler" (65 nm) processor, is this possible to add support for it or at least share source code (win32) so i could add support myself ?

Regards
BMN

 

The Tyler core works with RMClock which gives you more options.

@ kleewho, i had all the basic options covered, plus i was working on a way to get more pstates, like RMClock does for the Tyler and older cores. But it's tricky and i don't know exactly if it works fast enough to be of any real use.

 

As Th3_uN1Qu3 said, older chips like Turion64 are supported very well by rmclock.
Anyway, sourcecode is available in the linux folder inside tpc rar archive.

 

I've Turion 64 X2 TK-53 and it doesn't work with RMClock very well, ie. when i set multiplier to 4,5x and voltage to 0,800V after some stress test it hangs up, but when i set the same voltage and multiplier under crystalcpuid it's rock stable. I suspect it's something with northbridge voltage, timings or something like that. I wanted to run "TurionPowerControl -l" to read northbridge voltage and other parameteres set by crystalcpuid and rmclock and find where's the problem.

@blackshard83
Do you know any program capable of read all parameters like FID, VID, northbridge VID and timings for "Tyler" core? I've tried K10stat, pumastatectrl, turionpowercontrol but none of them works.

I've found this sourcecode under linux folder, but it's possible to compile it to work under windows 32bit and how to do that?

Sorry for my bad english, but it's not my native language.

Regards,
BMN

 

Hi notmebug, unfortunately k10stat, pumastatecontrol and turionpowercontrol are for newer chips (family 10h and family 11h)

Your Turion64 Tyler processor is family Fh processor, and actually it is a K8. It is strange that rmclock gives you problems, usually it works very well with all K8 processors like older Athlon64 and Athlon64 X2 processors. Unfortunately I can't remember other programs like rmclock that do the job you are want.
I'm not very used to Turion64, but once I had an Athlon64 X2 and it was clear that it was not a good idea to use multipliers below 5X because some motherboards don't like it at all.

To compile turionpower for windows, you need Microsoft Visual C++, that's freely available from microsoft in the Microsoft Visual Studio 2008 Express package. Anyway, compiled binaries are already available in the archive file I provide on this forum (x86 -> 32 bit, x64 -> 64 bit).

 

@blackshard83

Could you help me with compiling TurionPowerControl for windows ?
I've downloaded and installed Microsoft Visual Studio 2008 Express Edition, after that i ran Microsoft Visual C++ 2008 Express Edition, went "File/New/Project from Existing Code..." choosed "linux src" directory, set "Console application project" in Project type and clicked finish. I've downloaded WinRing0 and copied missing "OlsApi.h" "OlsDef.h" files to "linux src" directory, and also downloaded missing "unistd.h" from http://linux.die.net/include/unistd.h . But when i'm trying to build project i get "Cannot open include file: 'unistd.h': No such file or directory". any ideas ?

Regards,
BMN

 

Mmmh, it's strange you need unistd.h.
I don't remember I included such header in my sourcecode.
You just need the two header files coming from WinRing0 and linking the static library WinRing0.lib inside the Project Settings.

If you don't find the solution, I can pack and give you the whole visual studio project if you want.

 

Thanks blackshard83 for your invaluable help.
I've linked WinRing0.lib as you said, but still same error. I've googled but found nothing.
Could you just pack whole visual studio project and send me to [email protected] ? I would be grateful.

Regards
BMN

 

Hi

I put the project in the first page of the thread, so all people who is interested in can download it. I hope it works and compiles, since the software configurations are a bit messed up.

 

I started project page on launchpad. Your version of sources are avalaible here and if someone want to download it just have to install bazaar and do it like this:

bzr branch lp:tpc/stable

I hope the download will work, if not just let me know and i will try to figure it out.

The trunk branch will be in long term the GUI/CLI version with all features from your source. At this time it's just changing the FID, DID, VID. I'm almost sure it wouldn't work on windows for now. I will check it I hope soon.

If anyone want to help me with new Qt GUI version (there's also a bit work under the hood) just email me.

There's also package for ubuntu avalaible in my ppa. It's from blackshard's original sources. Trunk version will be avalaible as a package later.

I will upload binary version for windows as soon as I learn how to do this on launchpad site

 

Hi there, Look I read the post and downloaded the program and made it to run, Still i don't get it, I am trying to under volt my laptop since i think it will free me away from this heat problem I have. Is there a way you can guide me into the use of the program please...I need this for a HPdv4 with Turions...
I will appreciate this and please never mind my lack of knowledge...please forgive that. My fan is running too fast and too much heat on my drive and it shuts the system down...it is abnormally heating...
maybe if you can help me with typeing a set of values that can under-volt or set me in a friendlyer setting for my laptop?

 

First write down default values. So simply run:

tpc -l

Then you can change the FID for active pstates.

You do it like this:

tpc -pallc (pStateId) (vidId) (didId) (fidId)

in my case it is:

tpc -pallc 0 11 0 44
tpc -pallc 1 11 1 52

I've got only 2 power states on my athlon. If your notebook shutting down then you might consider cleaning the fan and radiator in it. You could download service guide for your HP.

@blackshard:

I found that setting HTC temperature limit and hysteresis works, but the actual lower temperature limit (HTClim - HTChyst, but i'm sure you know that) seems to be a little higher then the value from this simple equation. Is it behaves the same on your system? I tried to figure it out, readed the manual for 780G (found it hard to understand the definition of the Tctl) and readed your source code but have no clue.

 

Well on my system it behaves in a bit strange way.

My htc limit is set by default at 100 °C and htc hysteresis is set at 95 °C (100°C - 5°C). Anyway it is disabled by default.

Once I enabled it and tried to set htc limit to 70 °C (so htc hysteresis went down to 65 °C). It worked: when the processor was reaching 70°C, it went to the HTC pstate (800 Mhz, 1.200v), but hysteresis was not respected at all. As soon as the processor goes below 70°C, it went to pstate 0 (I was stress testing it with prime95).
So I got a fast and continous jump between HTC Pstate and PState0.

You can find some more infos if you check the Family 11h Bios and Kernel development guide. It is freely available documentation from AMD site.

 

So how's the GUI going? I may finally get out of my laziness and get more pstates for us Athlon guys. That is if i'm still able to figure out C++, i never really learned it properly.

@ blackshard, how fast can TurionPowerControl process a command? Displaying all that text takes a while (a lot more than would be needed for switching pstates on the fly), but is the actual operation faster? I have a few ideas on how to do it, if i can translate them to code it would be perfect.

 

@blackshard
in here hysteresis works, except the wrong values. I've got the Kernel and bios guide. It helps me to understand your code because it lacks comments/documentation Did you tested changeing the frequency. I mean is it really works? Do you see the drop in performance if you set much lower freq? I have problem with this, but maybe it's my cpufreq-utils change it to default values.

@Th3_uN1Qu3
You could try GUI from http://launchpad.net/tpc. Just download it from bzr, compile and try. I still don't have windows and I don't have much time to install it, so it would be great if someone else could try it. It needs qt installed, and also qwt since I'm plotting the CPU temperature and HTC limits. Installing qwt is not so straightforward like other software on windows, it's not hard but you need to use make. You should be able to change the pstates characteristics (fid, vid, did) and play a little with HTC. I'm trying to tide up the existing code for now, but I'm going to add the rest features in some time.

How do you want to add PStates? By constatly monitoring the load of cpu, and changing freq and vcore? I could give you permissions to commiting to bzr branch (don't have much experience with it, so it would be a harsh on start). I'm sure that you can have your own branch so you could test your code, and if it would be stable and working it could be merged together with the rest of code.

 

Ok i'll try. I know how to use make, don't worry, BUT GUI-oriented people generally don't so we'll have to devise an alternative.

And yes i intend to change frequency and voltage on the fly. I don't know how fast it can be done however, nor how much processor time it actually uses. If it's slow or resource-demanding, it's not worth doing.

The problem is that i'm very very busy right now and in a fair amount of trouble as well - i didn't even have time to update my own QuickPlay Remapper. The 1/4 done new version has been sitting on my hard drive for more than a month.

Also i have some other hobby projects going, mostly electronics related. Except for my internet browser, i've been spending all my computer time in LTSpice lately. So anything to do with coding has to wait at the moment... I like coding, but building actual physical devices is far more fun.

 

I dont think there need to be an alternative since the GUI-oriented people generally dont make at all. All they need is binary file, so for windows it should only have to be compiled with static libraries and should be fine. On linux there's no need for static compilation, but still GUI-oriented people doesn't have to make anything.

It could be another program as well with only necessary functions - just a little daemon. How do you read CPU load? Is it OS specific? If so I would have to find something on linux

Sure thing

 

Yes I tried and as far as I remember it was switching to 800 Mhz and actually benchmarking reports a drop in performance. I don't understand why documentation says that HTC Pstate should be set to 1, while actually it is a completely different state with different frequency and voltage.

cpufreq-utils don't read the hardware registers (but they claim it), so they always report incorrect results.
You can use TurionPowerControl -CM to costantly monitor the processor pstate transitions (including transitions to pstate 6,7 and htc state).

 

Well I was trying to write a built-in scaler to let the program scale the pstate itself, but actually I wasn't able to find a valid hardware-based procedure to detect idle cycles.
I found that the processor has a profile counter event that could be interesting (EventCounter 076h, see the BKDG documentation guide), but had no time to check it out and had no real desire to understand the mechanism

 

Hmm, if YOU had no real desire to understand the mechanism, well sure as hell it won't be ME desperately trying to understand it. For the time being i just think i need another beer.

Anyway, you have no idea how much you've helped me to keep my shiny DV5 from melting. So i should at least do something in return but it's quite hard. Anyway when i'll be able to do something important for your program, i will. Right now the fridge will "do" another beer for me, for anything else, we'll see.

 

Hhehe, then have a beer for me too!

BTW, I just worked the thing out, and I have a perfect hardware based (and completely portable!) cpu usage counter I was right about performance counter 0x76, it counts the non-idle cycles of a core/processor.

 

Okay that won't be a problem, unless you have to go to school tomorrow (which i have, but anyway beer is better than school ). So the 0x76 register counts any "load" used cycles? That would come in pretty handy. I'll help myself to your source code and answer the rest of the questions myself. That is of course i'll have time in this overwhelming mess that is my life.

 

Great news about that 0x76 counter and thx for the tip with -CM. I'll have to figure out how to show this info in GUI version.

If you're writing something maybe you could try the version control system on launchpad? The code is already there

 

Another idea popped up in my head - why don't we patch into the OS power management table itself and force more pstates there, letting the OS do load management? It seems the most logical approach to me. AFAIK it's quite easy to do on Linux, Mr. Bill's OS will be more problematic here i think.

What do you think, is it doable? Anyway, i think i need to take a peek and see how RMClock manages pstates, i think our answer is in there.

 

Well that was my first idea, and I think it's the best way to deal with power management and pstate switching, since there is less overhead about privileged instructions. There's also the C1E state that is said it needs some magic from the kernel...

Then I realized I had to deal with the kernel and decided for a separate application.

About more pstates, there are three ways:

1) Use the available slots in pstate table (I think at least pstate 3 and 4 are usable)

2) Use the pstate control to change pstate on the fly. It looks like (but I haven't verified) that there is a MSR that allows to change FID/DID/VID directly instead of switching to a power state.

3) Always use one pstate and program it as necessary, and then issue a transition to itself to "refresh".

I'm not sure which method rmclock was using, maybe method 2, but I think the safest to follow is method 1.

 

Well that would be perfect, but how do you convince the OS to switch to them? That's my point, using your software i could switch to whatever pstate i felt like dropping my settings into, but how to make the OS do that by itself?

 

Sorry for being away for a while, just being busy a lot.
Well, we can just disable OS scaler and allow tpc to switch pstate itself. I'm planning to add a command line switch to allow automatic pstate management based on a configuration file.

Glorious rmclock works exactly this way too.

 

ah! I'm pleased to announce I got a nice and working cpu scaling solution and a nice configuration file support

As soon as possible I will upload everything here, but unfortunately I have no Visual studio currently installed, so people with windows have to wait.

At the moment I'm happily testing the thing with linux x64 and it looks like pstate 3 is working like a charm

I don't know if pstate4 would be usable, but I'll investigate.

to kleewho: sorry for not using launchpad, I hope you want to deal with merging the projects