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

It's still incredible your processor runs at 2.0 Ghz with such voltage, can't believe those screenshots!

Anyway, I think that linux cpu frequency scalers could be modified with no excessive work, but i don't know it is a good idea since we're always talking about pushing the processor out of its specifications.

 

Well, it has been doing fine for a few weeks. There is an obvious difference in heat output and a 20 minute increase in battery life (on a measly 6-cell!), so it's most certainly working. I'm still curious about why it goes crazy if i edit pstate 6.

I've been overclocking since i was 13, and in 5 years of pushing crazy frequencies i have only fried one piece of hardware - a P4 Northwood CPU, that i ran for over an year pinmodded to 1.8v, and without adequate cooling. Poor thing was too slow anyway. I have also done voltmods on a few video cards and none have failed.

The biggest risk of undervolting is losing data, and i believe everyone doing it is aware of this and has backups. I just want to know the way to do it, besides, if no one tests it we'll never be able to know if it's safe or not.

 

Yes you are correct it is 550mhz don't know where I got 525mhz from (so it was 550mhz @0.7500v)...
I have been toying with K10STAT and some others as well but as a person who used Rmclock for many years most of these new undervolting tools just don't come close. I wish the makers or Rmclock would release the source code so development could continue.

 

Probably there will be no chances to see some improvements in rmclock, since the guy who was working on it quitted from rightmark. Actually it is a very long time it doesn't get updated, and probably it will slowly die by itself. It's a shame because it was kinda good software, and probably can't go opensource since it probably requires some nda agreements.

 

Well, I'm tweaking processors and other tweakable things since years too

BTW, another curiosity about pstate 6 and 7 is that when I reboot the system, they get a new configuration similarly to pstate 1, pstate 1 too get "restored" to its original state.

Also there's a boot pstate. That's the very next to default maximum pstate, so with ZM/RM is pstate 3 (psmax=2) and with QL is pstate 2 (psmax=1).

 

Yeah, tweaking things is fun. I picked it up from my dad, but took it to a larger extent that he did.

However, isn't it normal for the pstates to be restored upon reboot? The CPU is reinitialized on a reboot. I have discovered it is also reinitialized after suspend, and since i've fixed my network issues and now my wired card resumes correctly after suspend, i have also added the turionpowercontrol commands in /usr/lib/pm-utils/sleep.d I already had them enabled upon startup. I have also tweaked the ondemand governor to step up the CPU at 60% usage rather than 95% how Ubuntu Jaunty is configured, and dropped pstate 1 down a couple notches to 800MHz. Running very good so far.

So basically you say my CPU boots in pstate 2 (1GHz)? Now i know why the preinstalled Vista loaded so slowly... With Ubuntu the boot time is much better however.

Okay so... any chance of MORE PSTATES controllable by the OS?

 

For AMD processors the driver gets the table of available pstates from the BIOS/ACPI at boot time. Usually the BIOS creates this table dynamically based on the CPUID. However, it's possible to override the ACPI tables using a custom DSDT; I've done that on my Opteron 185 to enable C'n'Q (since the Asus BIOS didn't recognize the processor at all). Haven't tried it with this laptop yet.

 

Hmm, so what exactly is this DSDT and how do i use it? I've read a bit here but i didn't quite understand. Shed some light please?

 

Eh, never mind. The processor state table isn't present in my DSDT, it must be somewhere else.

Ah, it's in the SSDT. /sys/firmware/acpi/tables/SSDT to be exact. In fact the processor tables are the only thing in the SSDT. Now I just need to remember how to get it loaded at boot time...

 

Thanks for your effort. I assume i have to edit it with a hex editor right? Coz it doesn't look anything like a cfg file. Been a while since i hacked at stuff in raw hex.

 

Ok guys, some news here.
Doing some more test I found that if I lock the cpu scaler to a fixed frequency (any, I always prefer 525 Mhz) and then I issue:

TurionPowerControl -fo 0 0

core 0 will go into pstate 0 (as stated above), then it will automatically go into pstate 6 and flip flop between pstate 0 and pstate 6 on the base of the load!

If I idle, it will stay in pstate 6, if I run prime95 it will stay in pstate 0.

For un1qu3:

I see you have 1:30 hour of battery life at default and you're running linux. You should try to install AMD video card drivers (from your distro repositories, not from amd site) to allow powerplay function. It gave me 30 minutes more on battery life!

 

I have the latest ATi drivers from AMD's site and powerplay is enabled. The drivers that my distro's repositories provided were not the latest version and had lots of bugs. I don't want to go back to them, and battery life was the same with them anyway.

I'm saving for a 12-cell. Right now i just want more pstates...

 

I have the latest ATi drivers from AMD's site and powerplay is enabled. The drivers that my distro's repositories provided were not the latest version and had lots of bugs. I don't want to go back to them, and battery life was the same with them anyway. Btw, battery life was 1:30 on Vista as well.

I'm saving for a 12-cell. Right now i just want more pstates...

 

Hi, blackshard I was wondering where you went from the amd forums.
Anyway how about adding the ability to modify up(ms) and down(ms) to your program?
Also during Sleep Suspend CPU registers are resetted so actually adding it as a one time run method on log on doesn't really cover all situation.
Need to find a method to include running the program on computer unlock as well.
Also I noted that changing the FID has no implications on performance AMD could have implemented a multiplier lock on it.
Although the CPUZ shows a change in frequency there is no improvement in performance using a CPU burn in tool
Also the effect of NorthBridge Voltage is still relatively unknown, anyone shed some light?

Setting P-State 7 causes a crash while setting P-State 6 doesn't really works sometimes.
1.2V only occurs for a split second then it reverts to 1.075V
I read the Turion Documentation as well there is a P-State called S0.C0.HTC
and it is 1.2V (although there is nothing which says what it is supposed to do)

Blackshard program has a few advantage compared to K10STAT and vice versa.
1)Ability to adjust all p-state including the 6
2)Linux Support
K10STAT
1)Able to set throttle up down timings
2)Able to set processor to ganged/unganged mode
3)Able to set to run as a service
4)Easy config

Maxed I could achieve is
1.075V P-State 0
0.800V P-State 1
0.750V P-State 2

 

Well, sometimes I just check out amd forums, but there are always people asking about high temperatures and noone actually contibutes with infos.

This can't be done: my program just tell the processor to change its parameters, but actually transitions are requested by Operating System with its policy. In Linux you can change sampling period, up threshold, down threshold and so on, but Windows doesn't allow such modifications.

Well, it is not difficult at all to implement a function that just leave the program in background checking every 10 seconds if parameters are changed... Again, in Linux you can edit the sleep.d script file to call TurionPowerControl everytime it came back from suspend, but again I found no such viable way with Windows.

Yes, since the multiplier is locked in the up way, you can't overclock your processor. CPU-z just reads the same registers my program changes, so it gives a false information.
BTW you can lower your FID to obtain almost any frequency below your processor maximum frequency.

Yup, it is pretty unknown. Actually I see that the processor, when in halt state (c1 or c1e) consumes very very little power (ZM processors have a 300 mW TDP in c1e state, and 3W TDP in c1 state). Unfortunately it is very difficult to obtain some precise power reading on notebooks.

I noticed that if you set pstate7 with a vcore values that's below pstate0 vcore value, the processor crashes.
About pstate6, changing its settings always stop transition to pstate6 to me, but transitions to pstate7 still rarely happens.

The HTC state is Hardware Thermal Control state, where the processor enters if it goes too hot. It should be denoted also as the AltVID state in the documentation, but I'm not completely sure.

Yup, K10STAT is actually a program that controls the transitions of the processor: it substitutes the operating system default cpu frequency scaler (exactly as rmclock did). You have to uninstall c&q drivers (Windows XP) or disable the feature (Vista/7) if you want to let K10STAT control the processor frequency reliably.

I would like to implement such thing in my TurionPowerControl, but I want to find a very good way to detect cpu usage from processor performance registers (as rmclock did).

About ganged/unganged mode I think you mean the memory controller: there's a simple way to change ganged/unganged but hadn't tried to implement such thing because I wasn't really sure about the success of the operation during normal usage. Does it work?

 

I did some more testing with your Program and K10STAT.
The 1.2V issue can be fully resolved provided P-State 6 and 7 are overwritten with P-State 0 Settings. Works for me.
I think this is a AMD Issue or Microsoft Issue.
Does it replicate in Linux?
I think we can bring it up to AMD for errata.
Digg it or something.
We need to tell them P-State 6 and 7 is not disabled
Ganged/unganged mode works in K10Stat
Get a slight power boost in ganged mode 2

 

Switching to pstates 6 and 7 also occurs in Linux. And every time i would mess with them my laptop would crash a few minutes later. I will try to make them the same as pstate 0, but i doubt it'll do anything but crash again.

 

Ahhh I saw what does k10stat means for ganged/unganged!

Actually there's a sort of "bug" (it isn't a real bug, btw) which causes dropping in performances when both cores work at different speed.

Out turions (but also Athlon QLs and Phenom I too) have independent cores, so they can change their frequency independently on the base of the load.
On the other side, the Windows XP/Vista/7 process scheduler is a bit dummy because tries to spread the processes on the processor with lowest load (easy talking), so processes continuosly jump from a core to another. Look at this article from anandtech describing the problem with Phenom I:

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3344&p=3

So if you run a single threaded process you can lose even 10-20% of the processing power because windows scheduler is wasting time annoying the processor with useless transitions letting the process jumping around from a core to another.

The solution is: run all the cores at the same frequency! So here comes k10stat "ganged" mode.

For example, if I run super-pi benchmark (famous single threaded benchmark) I get the calculation of 1M PI decimals completed in 44 seconds if I set windows on High Performance. If I use Balanced profile the same calculation require 53 seconds, because Balanced profile allows frequency dropping while High Performance fix the frequency at maximum.

The same doesn't apply to Linux kernel, because I noted that single processes tends to be executed on the same core/processor, and this is more intelligent because it better exploits L1 and L2 caches.

About psate6 and 7, I just discovered that pstate 6 is invoked when a core is in pstate 0 and the operating system goes idling. It happens exactly in the same way both in Windows and in Linux, so I think it is a "feature" of the processor, yet an undocumented feature and with no apparent meaning. An AMD technician should tell us the reason of this choice.
Pstate7 is still a mystery.

 

Yea I attest to that when it is idling and processor is set to 100% it occurs.
I think I gonna live with it because of the power boost given by the ganged mode in K10STAT

 

You can use both tools in the same time: k10stat to deal with transitions and turionpowercontrol to tweak pstates 6 and 7, since turionpowercontrol works in "fire and forget" manner.

edit: BTW, that K10STAT ganged mode as a drawback anyway: it will consume more battery since it raises the clock of a core even when it is idling.

 

Well, the server roots of Linux are showing here. Too bad it can't do sh*t in the real desktop world. Thought i'd just chime in with some new results, but i may sound a bit incoherent since the gods of technology weren't too nice to me today.

I have decided to switch back to Windows and install Win7 RTM. Two reasons: FL Studio low latency (running such a program in a virtual environment is PAINFUL), and i want to play something else than UT2004 as that seems to be the only "real" game that actually plays nice with Linux. That's why i bought a laptop with a decent graphics chip after all, to be able to game some even when i'm not at home on my main computer.

I am very happy with Windows 7. Big plus no. 1: All special buttons work without the stupid laggy HP software that was in its preinstalled Vista Home Prem. Big plus no. 2: NO AUDIO DROPOUTS!!! I still can't believe it, but the DPC Latency Checker graph is all green.

Currently priming 2GHz at 0.9625v - Prime 64-bit immediately pointed that the vcore was a tad too low, and it just errored after 30 mins at 0.95v. Regardless, the difference between this 0.9625 and stock 1.1v is over 10C - now at 83C, while on stock volts the poor thing would go to 94C full load, and if i'm not mistaken 95C is the maximum design temperature.

So what's the bad you may ask? The partitioning program i used to shrink the Ubuntu partition and make room for Win7 (i intended for a dual boot for the moment) screwed up the Ubuntu partition entirely. A fsck from the live CD recovered everything BUT the home folder... Great. Eff that, wiped the drive clean again. Thankfully i only lost a couple beats that won't be too hard to make again, especially with ASIO working now. But still, it's 4AM, i have to be up early tomorrow as i'm leaving home, and my head feels like it's about to explode in a hundred pieces.

* Uses the nice Sticky Notes feature in win7 and types " Back up your DAMN WORK!!!" *

Thanks for the explanations. I'll play with my Athlon a bit more when my head won't feel anymore like it's about to explode. Any other interesting things you want to share, i'm all ears. Btw i still want MORE PSTATES.

 

I'm a huge fan of UT2K4, do you play online multiplayer?

Actually I found win7 being much much better than vista. There are some interesting optimizations here and there that finally make it an OS with a sense.

I managed to get over 100°C with wprime. That's not funny, becuase AMD documentation says that 100°C is maximum operating temperature, and I discovered that HP disabled (or did not enable, dunno) thermal protection!
Now I'm modifying my turionpowercontrol to manually enable thermal protection and allow to change parameters.

hehe, the home is the first folder which will fly away

Stay tuned for HTC protection enabling, about more pstates, well I need someone who helps with some code to obtain high precision core usage...

 

On my Acer HTC is enabled. There is throttling at >90 deg.
Then when it hits 100 it just power off automatically no a very pleasant surprise especially when you are gaming halfway.
After I did some intense hard modding it no longer hits 90.
I suggest hard modding be done if you are comfortable with opening up laptop.

 

I tried the linux binary compiled using g++
however when I run I enounter

Code:

cpuid:pread: Invalid argument
Cpuid_Fn8000_0008 Instruction failed
Error: unable to get processor specifications

I loaded the cpuid and the msr modules & CPU PCI configuration registers all exist so what could the problem be?
I ran using su
Thanks

 

Yes i do, but i don't have it legit so i play with a few classmates on a private server. Heh, back in the day there used to be this little patch allowing you to play on ALL servers... Wonder if it still works. But anyway, i don't think illegal software is allowed here, so that's enough.

Ditto. It - works - period

The only drawback i have found so far is the much lower performance on HD video. I could downclock my CPU to 400MHz in Ubuntu and it would still play HD 720p without a hitch since it was all done in the GPU, now i can't set the "power saver" profile without my video stuttering, thus i did not manage to see an entire movie on my way back home. Hmm, wait, there's a 64-bit version of the K-Lite codec pack, maybe that will help. Oh and MPlayer also is available for Windows, why don't i try that.

If your chip gets over 100C then i strongly believe that the cooler wasn't mounted properly at the factory, or it's full of dust. If it's clean, contact HP and demand a replacement. I took mine apart myself to fix the touchpad (i believe you noticed the thread), but i did not have a warranty as i didn't buy it "officially". It was brand new unopened, it came with everything (vista license included) except a warranty.

Btw, here's my priming result.

So i'm guessing you want to hijack the OS's power management (assuming it's Windows we're talking now) ?

@ weinter: One of the modules isn't loaded properly. I got the same when i was missing the cpuid module.

 

Sorry weinter, but I can be of no use with these modules and patches with linux. I just found that my fedora core 10 linux exposes the correct filesystem objects out of the box. I don't remember the path where they are exposes, but you can check the readme text file attached to the turionpowercontrol archive: if those filesystem object are not available, the program can't obtain cpuid/msr data.
Probably some kernel modules aren't loaded correctly, I suggest you to use the package manager shipped with your distribution to install them.

edit: thank you about your infos on thermal status. k10stat showed me that HTCEnable bit was off, then I found by myself that it was really off and managed to get it on. Now HTC works also on my laptop just if I undervolt PState 1. That's a bit of shame

 

Ah! Well I paid 10 euros two years ago for that I think you can find some nice offer on steam, if you're interested.

You can also try Media Player Classic Home Cinema. It is working good for me, both cpus are always at 525 Mhz even with 720p videos.

eck, the problem is that I have warranty and don't want to lose it opening the notebook by myself. The heatspreader is clean and I have enough technical skills to open the notebook, remove the cooler, apply some new thermal paste, clean the heatspreader and then close it and have a happy life. The dilemma is: does a happy life worth losing my warranty?

I just sent a mail to HP, they told me some useless things to do, now they are asking me to restore the software configuration and send them the notebook. Then I'll have to wait three to infinite weeks to get it back for something I can do by myself and probably I can do much better. :cry:

Nice prime
I just want to let turionpowercontrol handle transitions on the base of core usage. I just need a way to get core usage, then transitions are easy to do. No need to hijack os' power management, users will just need to turn it off manually (like they do with k10stat)

 

If you just override the ACPI tables then the OS power management will take care of everything...

 

Mmmh, don't undestand, are you talking about thermal control?

 

No, he's talking about the whole ACPI configuration. They list pretty much every power management spec of the machine and make it available to the OS. He explained a bit earlier that the CPU states are stored in the table called SSDT.

The ACPI tables can be edited directly under Linux. In Windows you can fetch them by using the GetSystemFirmwareTable function. I would've played with it myself but i am extremely busy right now, gotta go.

 

I get a funky error when trying to run it on my DV9700. Any ideas? It's a Turion TL-60.

 

TL-60 is Tyler Chip.
BlackShard's program works for Griffin Chip.
You have to use RM Clock, different power design.

 

Thanks, i guess i figured it out myself. Btw, RMClock did give me MORE PSTATES. 7 to be precise. Any integer multi from 4x to 10x. So i'm guessing it edits ACPI tables to do that?

I set my voltages as follows: 0.7625 @ 800MHz (4x), 0.8250 @ 1200MHz (6x), 0.9000 @ 1600MHz (8x) and 0.9500 @ 2000MHz. The rest were automatically derived by RMClock. In this configuration it is IntelBurnTest stable. Though i forgot to check the temp improvement... Before temps are here, i guess i'll run it again to see how it does. But one thing i know - this thing idles really really cool. I left it running overnight with some downloads and i couldn't hear the fan at all till i fell asleep, and now when i woke up the fan is still running at minimum speed and the CPU is at 45C. My DV5 gets pretty hot to the touch if i leave it overnight, and the fan always spins at medium to high speed making quite a bit of noise.

Edit: Results here. Nothing short of awesome. Fan never sped up to more than a faint hum. Before undervolting it would occasionally kick in to the mid-high speed that the DV5 runs at almost all the time... Now it never got past the second speed level. If i'm not mistaken this fan has 4 speeds.

 

Actually I don't know if rmclock uses predefined processor slots or maybe programs the transition on-the-fly, letting you use all the multipliers available. Actually C&Q features are pretty different between old Athlons and newer ones, mostly because there is a new memory controller and a completely different approach.

 

The new athlons has dual power plane able to set to unganged mode the old ones are pretty much ganged only.

 

I don't think it programs it on-the-fly, because it shows me the list of p-states and i can check/uncheck as many of them as i want, and set the VID on any of them. The only thing that didn't work were 0.5x increments on the multi, it only works with integer numbers. But 7 pstates is enough already.

 

Well I think it's exactly what I meant.
Once I had an Athlon X2 BE-2350 (desktop) and rmclock gave me a lot of of pstates, 13 if I remember correctly (from 4x to 10.5x, with 0.5x steps) and I had the possibility to enable all of them. Actually older Turion/athlon processors are pretty different from our Griffin processors, so I can't be really sure about the way rmclock works.
BTW it is possible to program our Griffin processors on-the-fly and obtain possibly infinite pstates Just use one pstate (for example, pstate 0), then reprogram it with different FID/VID/DID when needed (for example, when cpu load goes up or down) and then force a transition to the same pstate. That should be flawlessy and give you all the pstate combinations you want

 

Just wondering, would you like me to do a GUI for your program? It would be less confusing for people who have never seen a command prompt before, and would give me an opportunity to practice my C++ skills.

 

You're welcome
I would deal by myself, but actually I'm quite busy. Also I would like to try something with Visual Basic 6.0 or Vb.NET, since dealing GUIs in C++ is a bit painful in my experience.
I found VB5 and VB6 are easiest for GUI shaping, but also Delphi is good enough.

The big problem is that designing native interfaces is in contrast with portability :/

 

I hate VB, it's so slow and buggy. I heard that Delphi is nice but i don't feel like learning it... As for portability, well, Linux users are more confortable with the command line so i doubt the lack of a GUI will bother them. I could do it in Python as that would be more portable, but again, it is too slow for my liking.

 

Yeah I know VB is slow, but for interfaces is truly helpful and fast. Also the task is not cpu intensive at all, so maybe it worth consideration. There's also the .NET platform: it is quite simple for interfaces (as VB is), but unfortunately needs the .NET elephant runtime.

 

After struggling with Dev-Cpp and wxDev-Cpp that were *very* buggy, i finally decided upon AutoIt. I have most of the actual code done already, now all i have to do is map it to some pretty buttons.

 

Newer version of TurionPowerControl v0.12

Available here:

http://amdath800.dyndns.org/amd/tpc-0.12.rar

New features:

- Hardware thermal control (enable/disable, parameters check, etc...)
- More general status informations
- Voltage slamming control
- Check mode (-CM undocumented parameter) offers some more real time infos.

Launch the program without any parameter to get helpful description about available parameters.

 

Very nice.
I am now trying it out because somehow windows 7 load WinRing0 much much faster.
Btw are you into SLIC BIOS modding ?
PM me

I think you need a more comprehensive guide too much options even more than K10STAT

 

I was just a fellow of the thread, but can't really figure out how to manipulate a bios. It would be kinda really nice anyway. I just discovered on my desktop that disabling Parallel ATA controller speeds up the boot process a lot. Unfortunately can't do that on my notebook and curiously HP left Parallel ATA controller activated on dv5s.

About the guide, you are right, the program has grown and many options aren't described anywhere. Maybe I will update the readme inside the program later. If you want you may also write a guide for it, I'll be happy to ship it with the program.

 

No SLIC BIOS modding is to activate Windows by the OEM way.
SLIC 2.0 is used to activate Windows Vista your BIOS should have it.
The question is to update SLIC 2.0 to 2.1, you should be able to guess what 2.1 does.

 

The optical drive of the DV5 is on parallel ATA.

 

Ahhh, got it... I was misunderstood, excuse me!
Now I just remember the fact about licensing, I read it somewhere but actually never digged too much about it. I'll try learn something if I have some spare time.

 

Mine isn't. I disabled the Parallel ATA controller in Windows 7 and the optical is working.
During the boot, there is a someway long pause during the first part of the booting process.
I have the pause on my dv5 and had the same pause (but just little shorter) on my desktop intel machine. Both of them have no parallel ata devices.

I disabled the parallel ata controller in bios in the intel desktop machine, and magically the long pause has completely gone.

Unfortunately can't do the same on the dv5, but disabling the device in the control panel gave me a little shorter pause (measured watching the boot animation of windows ).

edit: mmh, I just wondered that if your optical is Parallel ATA, you may not experience the pause during windows booting. I mean, probably the pause is due to the fact that windows is querying the IDE bus, and since there are no devices, it just waits for a timeout period because of no replies.

 

You shouldn't have this problem in Windows 7 as it loads driver in parallel to speed up booting significantly.