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

Maybe the problem is in another place: Turion processor can clock their cores separately to obtain some more battery life. This fact is good for power consumption, but is bad for performances because Windows process scheduler is weak in this sense. It is built on the fact that both cores/processors must run at the same speed, so it will just make the video player process jump from a fast core to a slow one and so on.

You may try to force affinity to one core on the process using Task Manager, Right Click on the process and pick Set Affinity... and see if the movie still stutters. BTW you can use the command line I gave you exactly as it is. Your processor is exactly the same as mine, but mine is absolutely the most unfortunate I've ever seen, so my undervolting settings are some way "conservative". They should work kinda well for you. If you see sudden crashes or blue screens, then your processor is less fortunate than mine

 

Yeah, I've been trying out your command line and it seems to be working very nicely so far....don't know why but without a doubt XBMC plays my movies waaaay better with the windows power management disabled (cpu min. power 100%).

But that with your command line seems to be working really well for me for now - fingers crossed!

So I can just create a .bat file for that and run it automatically at startup right? (my pc never sleeps or hibernates)

Any need to run KSTAT10 as well?

Thanks so much for all your kind help

-------

oh...one more thing....your command line uses command "TurionPowerControlx64" whereas I can only get it to work using cd x64 and then command "TurionPowerControl". (using the downlad on the first page of this thread. I presume that's what you meant?

 

Exactly.

Not if you don't need KSTAT10's features. Personally, i don't use it.

Yes.

 

Thanks so much for the help

I started tweaking in KSTAT10 as I could use Weinter's guide to feel more confident.

Intel burn test hasn't finished yet but after 3 passes it looks like I'm running stable at 1.0000V for Pstate 0 !

----edit just finished - stable at 1.0000V after standard test ! I'm so happy ! -----------------

I'm also running at only 68 degrees during the burn test when previously I was idling in the mid 80s!

So...

A couple more Questions, and then maybe I'll finally be able to leave you in peace!

a) Whatever result I get from KSTAT10, how would I slot it into use with turionpowercontrol, and what (if any) would be the advantages of doing so?

b) If I leave Windows min CPU power at 100%, will the Pstate changes governed by turionpowercontrol or KSTAT10 still happen?

And if they do, is there any harm in leaving it in Pstate 0 all the time if it makes me happy?

c) If so, would changing clock control to Ganged (highest) make no difference, or could it still improve performance? Is there an equivalent function in turionpowercontrol?


Phew, this has been quite a learning curve - but very fun! Thanks so much for your patience with me guys

 

Consider yourself very lucky. My ZM80 won't go below 1.0875v and reaches almost 100°C during stress tests.

to set PState 0, frequency 2100 Mhz (FID=13, DID=0) and voltage 1.000v (VID=44) you have to launch this command:

TurionPowerControl -pallc 0 44 0 13

There's no harm in leavin it in pstate 0, or at least there should be no harm. If you computer shuts down, then there's a problem as I said before. But in theory there's no harm in leaving the processor at maximum frequency.

TurionPowerControl doesn't govern pstate changes, it is just useful to set pstate parameters, like frequency and voltage.
k10stat is useful to change pstate parameters and is also able to change pstates.

If you want to use k10stat to change pstates, you have to inhibit Windows pstates governor. Setting bot min and max cpu to 100% in Windows is a good way to inhibit the Windows governor and then reliably use k10stat governor.

Using ganged in k10stat is a way to improve single-threaded performance, but will likely produce more heat and less battery life than unganged (that's the default method Windows also uses).

TurionPowerControl has no such feature because it does not govern pstate changes.

No problem

 

Well, I guess I wasn't quite as luck as I thought....even though I passed the stress test at 1.000V for P0 I think K10stat and Vista power options were conflicting with eachother and the stress test was done all under P2 or P1 !

Anyway...I think my number is 1.0375 which is still pretty good


I am now back to using TPC, as it is simpler, and K10stat seemed to fluctuate between Pstates even when I told it not to.

However, even TPC seems to have some issues if Vista's Power options are set to "High Performance" (100% min CPU).

If I run Turionpower control cmd with this setting in Vista Power options, nothing happens! (P0 voltage stays at 1.1V)

However if run my turionpowercontrol command with balanced power settings, the changes do take place. Then if i change back to "High performance" in Vista my new P0 voltage is seen.

So...I am able to get the desired result, but not in a way that can be run as a batch file as startup.

Any reason you know why TPC is not "taking" with Vista Power options set to 100% min CPU power?


Thanks!

edit 1:
even weirder....even after using the above process to set P0 constantly, (so P0 is active at idle), during Intel burn test I am seeing some drops into P1 ???

I am using CoreTemp64 program to monitor Pstates and temp.

edit 2:
Could one solution be to overwirte pallc 1 and 2 values with pallc 0 settings? Or is that a bad idea for some reason?

 

If the CPU is in the pstate you are changing values for, changes are not applied until the CPU switches to a different pstate then back to the one you changed.

So a simple solution to this would be forcing a pstate switch in TPC after you give the undervolting commands. Read the readme, it has such a function too.

 

Are you sure it is dropping to P1 and not to P6 or P7? By default, P6 and P7 have exactly the same frequency P1 has, but vcore is 1.200v.

I don't think it's a good idea. You can use TPC switch -psmax 0 if you want the processor not to reach any pstate over 0.

 

Hello Guys!

NPC is a nice program. Nevertheless I am facing some problems. I don't know what the reason could be.

I wanted to use a power saving mode, so I set TPC up for:

Pstate 0: 1500Mhz, 1.0875V
Pstate 1: 1000Mhz, 0.7750V
Pstate 2: 500 Mhz, 0.750V
-> works well

I confirmed running 1500Mhz at Pstate 1 with 0.9V easily beforehand on my normal configuration.

Attempting to use the value 0.9V at Pstate 0 gets an BSOD immediatly. Also everything below 1.05V gets me that BSOD. What a .

Anybody the same problem? Any suggestions?

 

Well, now that I've finally fully figured out what I'm doing 1.050 seems to be my stable voltage!

It definitely helps, although I'm still running a little hotter than I'd like. I remember a few pages back one of you mentioned dropping the P0 frequency to 1800Mhz.

Would I really not notice any performance hit from that? (Again specifically for playback of HD video).

Any reason for 1800Mhz, or could I try 1900 or 2000? (Or are those not enough of a drop from 2.1Ghz to make a significant temp diff?)

Thanks as always

 

Are you sure you weren't just exceeding default FID value?
I mean, if your default FID value is 12 and you set 13, your processor will still run with 12. Software like tpc, cpu-z, coretemp, etc... will say that you are overclocking, but the benchmarks will tell you the truth.

If you are not exceeding FID, but just using pstate 1 to set 1500 Mhz using valid FID and DID combinations, I have to say I noticed by myself that setting pstate 0 to 1100 Mhz require a lot more voltage than pstate 1 set to 1050 Mhz.
Actually I can't really explain the reason why, but it would be nice to do some experiments this way.

 

So what combinations are valid?

I simply set -pallc 0 XX 0 7 for 1500mhz at p state 0.

 

Well, drop in performance is actually almost proportional.
AFAIK a 2100 Mhz dual core processor is absolutely enough for any HD video, except if you use tons of postprocessing effects.
A 1800 Mhz dual core processor is almost the same for HD videos, I think there will be no differences at all, but I'm still crippled with the fact you need such high frequency to get a HD video running.
I was almost able to run a 720p HD video on an old single core 1.5 Ghz Pentium M using CoreAVC codec (which I suggest to you if you use software decoding).
Now, using hardware decoding built in my hd3450 card, the same 720p video can be easily decoded with both my processor cores at 525 Mhz.

Why don't you try MediaPlayerClasssic Home Cinema (and be careful to get the "home cinema" version), follow the simple instructions to let hardware hd video playing with low cpu usage and see if you have stuttering?

 

Yeah, I know I can get 720p running with much lower frequency, especially with COREAVC and no deblocking.

My issue is
a) I'm hypersensitive to the tiniest bit of jerkiness in panning etc.
b) I want to use XBMC which is a very CPU intensive player for HD content.

But I think I've got it all working fine now. I tried compressed air in all the vents again yesterday, and this time it really seems to have made a difference! So, coupled with the undervolting, touch wood, my overheating probs may be behind me

You (and th3 Un1q3) have been so kind and patient helping me through all this

 

Any combination with FID less or equal to your default (=maximum) FID is valid. The combination you used above is valid, for example.

An invalid combination would be:

-pallc 0 xx 1 22

where fid=22 and did=1. This is invalid because fid is too high and probably exceeds your default fid (which is 16 for 2.4 Ghz processors, 15 for 2.3 Ghz, 14 for 2.2 Ghz, 13 for 2.1 Ghz, 12 for 2.0 Ghz and 11 for 1.9 Ghz).
The value wil be accepeted by the processor, but it will run at the default fid.

Anyway, you talked about voltage discrepancies running 1.5 Ghz at pstate 0 and pstate 1, can you give me more details please? It would be very interesting.

 

You should check if XBMC can be instructed to use external decoders like coreavc. Usually mediaplayers and mediacenters don't come with preinstalled or builtin decoders for licensing reasons.
BTW I'm happy to see that most of the problems come from dust inside the computer. Very often it is the first and most common problem causing overheat.

 

Well, I don't have a clue about why P-state 0 would need more voltage than P-state 1.

Nevertheless I also tested my maximum frequency of 2300Mhz (ZM 84, HP tx2) at P state 1. In this case it will freeze at the same voltage as in P-state 0. (just as it should be) On the other hand lowered frequencies will get me different voltages in different P-states. (Only tested 1500Mhz so far)

I thought about solving the problem with using more P-states. (like you have described in a former post) But how can I deactivate P-state 0 in K10stat or TPC? I know it can be done with Windows (setting it to 75% max for example), but Windows won't support higher P-state number than 2.

Is there a possibility to recognize Windows of higher P-states or cancelling switching to P-state 0 with K10stat? Would be very happy if you can help me out.



EDIT: Got an idea! I could use TPC with the function di! Will test it.
EDIT2: No use, even when using di, K10stat still recognizes Pstate0, as a result it will switch to it...

EDIT3: Now I know why sometimes my PC instantly freezes. I am not able to use higher DIDs than originally included at the P-states. As a result it is not possible to get P-state1 higher as 1150Mhz...

Nevertheless it is not possible to lower VIDs in Pstate0 as much as in 1&2. But I am sure my frequency in Pstate 0 is as low as in the state I tested it against. (it simply feels slower)

 

Well, to prevent using Pstate0 you can simply tell Windows or Linux not to use the processor at 100%. In Windows, for example, you can set Maximum CPU Frequency to 50% in the power management panel. This means that it won't go over pstate 1 because Windows uses default frequencies coming from the processor and pstate 1 at default is half the maximum frequency.
In Linux you can use cpufreq-set command line tool that does exactly the same thing.

About the edit3, I had no problem using DID 0 at pstate 1. I was able to run the system fine with 4 pstates and pstate 1 was set to 1600 Mhz and 1.050 volts.
I think I have a different explanation about the freezing system: Windows uses a "rocket" approach when it has to raise frequency clock (eg: jumps directly to highest pstate) and a "ladder" approach when it has to lower frequency clock (eg: reduces clock step-by-step).
This means that pstate 1 (in a processor with 3 pstates) isn't used so often and also it is always used while the voltage is falling from the highest pstate.
When the processor is going from pstate0 (faster) to pstate1 (slower), voltage and frequency are falling. Since frequency go down faster than voltage, there's no stabilization time because while the processor is already working at pstate1 frequency, the voltage is still falling down. This could be explanatory:



So *apparently* you can use lower voltages (at same frequencies) for pstate1 than pstate0: that's just a matter of timing.

On the opposite side (but just to be complete), when going from a slower pstate to a faster pstate, there's some need for stabilization time: voltage is sent up, then there's a wait, then frequency goes up too.

edit: the really strange fact is that using DID 0 requires a lot more voltage for the processor, and it is independent from the working frequency. I did some tests and found that my processor is stable with 0.7750 volts at 1.050 Mhz (FID=13, DID=1) and requires 1.000 volts at 1.200 Mhz (FID=4, DID=0). That's a huge jump for just 150 Mhz and this is really puzzling.

 

I am sorry to tell you that the Voltage / Frequency explanation is not the point.
Simple reason: Frequency and Voltage are pinned when I test stability, so the system is in steady state.

About simply setting Windows to 50%: Jeah, thats right, but Windows doesnt recognize Pstates higher than 2. So I wont be able to use Pstate3 with windows. I will need K10stat to do that. But K10stat doesnt allow 50% like Windows.

Nevertheless, seems like its no use anyway, since you figured out that it is not pstate0 but did0.

Thanks for the information about that DID difference. Haven't tested Pstate 0 with Did1, maybe this will work...

EDIT: Awsome! You figured it out! I can set my DID0 to DID1. So my Power saving mode is working much better now Thanks a lot!

Its sad that for frequencies above 1150Mhz DID0 is needed which is increasing voltage dramatically.

 

Blackshard in case you want some info regarding the new Turion II Chip.
Some users in this forum already has Turion II CPU.

 

Very interesting! Thanks weinter, I'll subscript the thread

 

I don't know why that guy is complaining about a max temp. of 58C, tbh. I wish older AMD processors were that efficient as my max is 76C

 

Well, he's not complaining about 58°C max, but about the noisy fan.
I had the chance to deal with an acer 5536g with athlon ql-64 and the fan was always on and spinning fast. It's definetely an issue acer has to correct with a newer bios.

 

Hello,

I was looking for that kind o software for about a year. This is great. If you would like to license it on GPL or MIT license I would try (never done this before ) to build a package for ubuntu with your software.

If you're interested just please let me know.

A note for linux users... I didn't find that information in this thread yet. To make TurionPowerControl working you need to have 2 kernel modules loaded: its msr and cpuid. You can check it with:

Code:

lsmod | grep -e msr -e cpuid

If these modules are not loaded there should be no output from this command. To load them you need to install them first with your package manager (e.g. with synaptic) and then:

Code:

sudo modprobe msr

or/and

Code:

sudo modprobe cpuid

It is working for me

 

Hi. Actually my program is "unlicensed". Consider it as free open software. I know nothing about GPL and MIT licenses, if you want to make a package of it, feel free to do that
About kernel modules, there are some indications in the readme.txt file attached to the archive and also there are some posts talking about missing modules in this thread.

 

blackshard , dose your software work with the neo athlon 1.6 L335 , if it is affected?

Thanks.

 

Not sure. Actually there are three architectures of AMD processors "in the wild" on the notebook segment: K8 (old athlons), K8L (newer 65nm Athlons and Turions, Griffin family) and K10 (newest 45nm Caspian family).
I'm not sure about, but Athlon L335 probably is one of the K8.

You can try using the old and good rmclock to tweak your processor, if it belongs to K8 family it will surely work flawlessy.
If if is a K8L (but I don't think so) you can use TurionPowerControl or k10stat. If it is a K10 there's k10stat.

 

Thank you for your response. I could write something about these license, but it would not be the whole true, because I'm the linux guy and we are freaks about the GPL

Yes, I know about the readme and the readme instructions told me what to load to get it working, but I really want to write that directly.

 

It looks like the NEO L335 is based on K8 architecture according to CPU-Z



Athlon Neo

Often compared to Intel's Atom, but probably more directly comparable to Intel's Consumer Ultra-Low Voltage (CULV) parts, the Athlon Neo is a low-power derivative of the K8 micro-architecture. AMD is pitching it for "ultrathin"/"thin and light" laptops to slot in around $700, above "netbooks" but below expensive laptops.

Conesus: Part of the Congo platform. Twice the cores of Huron. Some media outlets have suggested this to be a 45 nm part, but AMD's specifications indicate 65 nm. Possibly derived from Griffin, but certainly a K8 part.

http://www.10stripe.com/featured/cpu/amdcores.php

 

To be exact it is not the architecture but the power design.

 

I suppose when i get my notebook i will run blackshard83 software and see what happens.

 

Eheh, I can say you that it won't work, because the L335 is a Family Fh processor. My program instead works with Family 11h processors.
BTW you're not out of luck, because you can use rmclock (if you use windows) that's far more complete than my tpc or k10stat.

 

Blackshard83,

I'm getting high temps on my ZM-82 even in HP RECOMMENDED mode.
Even in that mode I get the intermittent spike to 1.2V. I thought it was only in high performance mode ,but on mine, I get the spike.

If I run 3DMARKS06, my CPU temp goes up to 90 degrees celsius still (HP RECOMMENDED mode).

HP offers me to RMA, do you think they have a way to solve the issue? (newer bIOS, firmware fix, fan override function...something?)

Thanks!

 

Hi exercise,

the spike always happens because it is independent from windows power management settings. It is just more evident when you use High Performance profile, but actually it always happens when there are some kind of power state transitions.

3dmark2006 really stresses both cpu and gpu, so your notebook is going to heat a lot and each component will heat the other, also because 3dmark06 is multithreaded and will push the cpu to heat a lot.

I don't think RMAing the notebook will solve your issue, actually I don't really know if it is a real issue too, because those temperatures are "normal" for HP technicians: once I had the chance to talk with one of them and they told me that if the notebook is really overheating, it will shutdown itself... Well, this is opinable IMHO, because the specifications of the processor say that its temperature must need go above 100°C and my processor reached 101°C under heavy stress with prime95 and linux x64 (it was undervolted too...).

Anyway you can try RMA it if you don't really need your notebook for daily usage.

 

Thanks for the reply!
I don't actually put a lot of faith in HP being able to solve the issue, just hoping they can offer me another computer like a Turion II 45nm of Intel C2D(although I strongly doubt they will). As I don't need my laptop for the next few months (I use it when I go away only), I'd like to take the risk. They obviously don't have a official fix (nothing in their download section, I'm running latest BIOS revision), they don't do COPPER Mod, they don't UNDERVOLT, so what do they do? If they know stuff we don't, and fix laptops only for people who complain, it might be it. Of, they just accept to look at it, reformat it and send it back to show they care?
I dunno anymore...I'm lost here.

Actually, sounds like AMD COULD do something about this...CPU driver or a revised Powernow that deals with the issue.
I mean really! If they don;t do it, it's probably just because they don't want to openly state that there is an issue/errata/whatever that cause the CPU to become overvolted (1.2V).

I tried to install/use your software, but I click on it, it executes itself in a split second and it's gone??? If I need to type a command line, where do I do it in Vista??? (there is no RUN option in start menu).

So true, one fan/heatsink combo for 2 heatpipe branches. 3Dmarks heats up both components, so basically, it stresses the thermal enveloppe of the computer. BUT if it heats up, what's the point of selling a laptop that CAN'T be used for what's it's meant to be...an ENTERTAINMENT platform! They have failed and they KNOW.
Maybe it's just me, I'm used to installnice expensive heatsinks on my desktop CPUs, Overclock them a lot and I never ever dealt with temperatures like that for CPUs...I can't believe this Turion is MEANT to stand 100 degrees celsius I DON'T.
The K8 CPU (desktop part) could NOT stand that much abuse, and IMHO, how could they adapt a mobile CPU to stand twice as much heat than a desktop CPU???

Choosing AMD/ATI platform (this time around) as the advantage of being cheaper and giving you more GFX horsepower right now. BUT if the designs aren't able to withstand the extra heat and the CPU voltage is out of control...well...

 

I now got K10STAT to work from startup and all.
The hard part was to figure out that you have to delete K10STAT.dat in order to register your changes, or they won't take effect AT STARTUP.
The only thing that doesn't work is that Pstate1 (50% CPU, 0.95V stock) is NOT changing. P0 and P2 are moving down, the way I set them. Surprisingly, the .95V@1100mhz doesn't move...
I got 0.762V@550mhz and 1.05V@2200mhz

Any suggestions?

 

It never happened to me.
I was always able to tweak all the settings and everything went fine with k10stat.

Are you able to change pstate1 using turionpowercontrol?

 

From where do you get the min and max values for VCore (or Vid)? TurionPowerControl shows me min 0.9V but I am able to set 0.75V for 950 MHz and everything is pretty stable (I did test it a little with phoronix test suite's stresscpu2 test). Could you explain this a little, I would like to show tests of your software on my blog, so I'd like to know a little more

 

Sure. Min and max come directly out of the processor.

You can check the "Family 11h Bios and Kernel developer's guide" at paragraph 2.4.1.3. There you can find the exact behaviour of MinVid and MaxVid values stored in MSR C001_0071 (that is: the processor won't let you go beyond its MinVid and its MaxVid values. If you go below MinVid, the processor will sent to the voltage regulator requests for MinVid voltage. If you go above MaxVid, the same applies). Unfortunately MinVid and MaxVid are read-only parameters, so cannot be changed.

 

K10STAT doesn't load at startup for some reason, I open it, click ok then the volts change. Pstate 1 NOW changes also.

So, for some reason, it seems to take a few times to make it work.
If I turn the laptop off and restart, K10stat will apply settings at startup. If I turn the computer off, then back on later during the day, back to regular voltages....strange...

 

I'm not really sure, but maybe you need to save the profile and recall during startup. Are you following weinter's guide?

 

So first, huge THANK YOU to Blackshard83. This program made my TX2 bearable (barely).

Now, I thought I should mention that I'm not sure if their is a bug, or running cores out of sync is a bad idea, but I have found that using the following command:

Code:

turionpowercontrol -pallc 6 28 1 15 -pallc 2 64 2 15 -pallc 1 62 1 15 -pall 0 0 38 0 15 -pall 1 0 62 1 15 -pall 0 7 38 0 15 -pall 1 7 62 1 15

Leaves me with a perfectly stable system which has a pstate 0 with core 0 running at [email protected] and core 1 running at [email protected]. I have used the intel burnin program to confirm stability.

My problem is, strange enough, executing TPC with this configuration causes some sort of slight lag in areo. I'm running windows 7, and the context menus etc collapse/maximize just slightly slower (noticeable). Interesting enough, other commands using tpc don't seem to cause any sort of slowdown for me.

I kinda like the diminished speed, I mean the windows look that much neater collapsing slightly slower BUT I know this is not how it was meant to work and so I would like to fix it. Any suggestions? I posted here directly because I'm pretty sure this is something to do with affinity since my cores are running at differnt speeds in pstate 0. Still, even with power saver mode on (where windows will often transition a single core up or down, or both) I don't notice the speed change. Only using tpc to modify the variables does it.

Should also add that I know it is not a speed issue because even outside of pstate0, after I modify the table, context menus are slow. AKA modifying just voltages and working in pstate 1 or 2 is fine. Once I modify pstate 0 though, even if the processor is only in pstate 1 or 2, there is a slowdown in areo. Hard to explain, but I hope you understand.

 

Sir! Yes SIR!
Must be a slight little detail that I miss or something. Right now, the app loads at startup, shows on screen, and I have to click a setting, click APPLY, then OK and it works. I reached 42 degrees celsius at idle once!!!!! playing cards or listening to music/email, it stays at around 50-55 degrees max (the ATI GFX chipset downclocks itself by 50% in 2D apps so it helps too).
At that temp, the fan is barely spinning, and the heat coming out the vent is barely noticeable.

I'M NOT SENDING THE LAPPY IN FOR REPAIR ANYMORE!!!

But I'd like to figure out how to make K10stat work automatic ,I'm getting there...slowly.
turioncontrol runs and disapear on my rig..dunno why (not 64 bits compatible maybe?)

 

Are you using windows 7? If it is a service issue, IE you can enter a chain of commands and use kstat10 without a dialog from a command prompt, then you can use scheduled tasks to handle it. Windows 7 tasks can be triggered by logging on, startup, and locking/unlocking an account. This means a task can cover all the bases for turion power control.

You will need to enable the admin account though, since tasks normally execute as a power user, not admin. Once you enable the admin account, you can set the task to run kstat10 with admin credentials.

Anyways I'm planning on posting a quick guide sometime in the next week or so. Still working on my TX2 atm.

 

MMmhh, I think I understood the behaviour, but I don't have an explanation for this, even because I'm not pretty sure about the entity of the "slowdown".
As you say, the problem happens just when you set something different to pstate0, and this is pretty unexplanable. Maybe, but just maybe, this can be related in some way to pstate6 and 7 too, but actually I can't really explain such kind of slowdown.
I'm sure about the fact that windows doesn't like different core speeds, because it assumes that they are all at the same frequency. This may create some nasty performance problems, expecially if you run single threaded apps.

 

Well turionpowercontrol just show the windows and then closes itself because it is a command line program. You should run it from command line to see its output, else it will execute, do its jobs and then terminate as soon as possible.
To let k10stat work in automatic, you should put some command line switches (they are explained in weinter's guide, aren't them?) to allow the program run and apply a specific profile and put itself in the traybar without doing all the tedious work yourself everytime.

 

Yes I realise that, it's probably just a mistake I do in the command line or an option that I didn't check. I just want profile 1 to load this is it!

 

Well, I found what is causing the slow down. Use a different DID for each core in pstate 0 and it causes it. I haven't tried the other pstates yet, but I assume it is just the use of a different DID for each core in the same pstate.

Interesting enough, this pstate doesn't have to be active for the effect: just the act of storing 2 DIDs for the same pstate causes the slowdown.

 

Yep, I tried and I discovered this issue too.
Anyway probably this is an issue with the timestamp counter. I think that on our processor the timestamp counter is calculated on the base of DID value to let it be invariant of the core frequency.

Probably Windows interface (and maybe directx counters too) uses the TSC to synchronize itself and toying with maximum pstate DID automatically break TSC normal behaviour.

 

well,my tx2 use zm-86,and it gets really hot when playing games.i try to change the pstate 0 from 1.125v to 1.1v,and it can safely pass through the cpu stressing test.but,strangely,it may crush occasionally when i normally use it.i don't know why.

core 0 pstate 0 - En:1 VID:34 FID:16 DID:0 Freq:2400 VCore: 1.1250
core 0 pstate 1 - En:1 VID:48 FID:16 DID:1 Freq:1200 VCore: 0.9500
core 0 pstate 2 - En:1 VID:60 FID:16 DID:2 Freq:600 VCore: 0.8000
core 0 pstate 3 - En:0 VID:48 FID:16 DID:1 Freq:1200 VCore: 0.9500
core 0 pstate 4 - En:0 VID:0 FID:0 DID:0 Freq:800 VCore: 1.5500
core 0 pstate 5 - En:0 VID:0 FID:0 DID:0 Freq:800 VCore: 1.5500
core 0 pstate 6 - En:0 VID:28 FID:16 DID:1 Freq:1200 VCore: 1.2000
core 0 pstate 7 - En:0 VID:28 FID:16 DID:1 Freq:1200 VCore: 1.2000