Mobile Pascal TDP Tweaker Update and Feedback Thread

If you use this tool to set the BIOS with the adjustable power limit option you CAN use a modified vBIOS to underclock via lower power limit, while still being able to have a high performance profile at default.

The "target" is the 100% power limit and the "limit" is the maximum allowed using the slider in Afterburner. The adjustable slider starts at 0 on the low end so can be used to severely underclock if desired. The stock desktop 151/170W that's loaded with the preset in the tool means a maximum of 112% on the slider.

I have my target set to 100W (100000mW) and limit 200W so in Afterburner I can adjust the cards anywhere from 0% (which instacrashed btw) to 200% and the % maps exactly to the power limit.

 

Yes, that would help me.

But for my little target / a little undervolt (<0,8v) / power limit this method is "much work". (order hardware and dissamble my notebook).

But this is the only option for me, right?

OR: I could use the battery-mode / gpu-behaviour in battery-mode (I think it is only a power-limit, 60% for example) for the ac-mode. But i don`t find a setting, for example "adaptive" in nvidia power settings isn`t a solution. In Windows Power Settings / Graphic processor set to "balanced" isn`t a solution, too.

 

If it bothers you so much why don't you sell your notebook and buy a 1060 version.

 

The best efficiency you can get out of your GPU in terms of performance per Watt, is to overclock to the limit of stability and reduce your Power Limit which I assume can be done using the tool in this thread. The overclock just ensures that you're using maximum stable frequency for each of the voltage points on the curve, and the reduced Power Limit will just give you the temperatures you're after - this way you don't need to mess with the Ctrl-F voltage curve in Afterburner, you'd just overclock as normal with an offset of say +200Mhz while having your Power Limit set lower.

 

I personally don't agree with what most people have said here, although it may indeed work, although this is more about how the curves work than power limits to begin with. But I could admit I am completely wrong if the power limit being reported is from curves and not TDP.

Reaching the power limit affects both framerate stability and FRAME TIMES (I don't know if MSI AB can show frametime, but I remember even Path of Exile's OSD could show frametimes). Whenever the GPU clock starts jumping around, frametimes get affected, and that just adds more stutters on top of an already cancer Operating System (along with whatever other bloatware is running).

What I suggest is:
Finding the overclock (point on the curve) where you do NOT reach the power limit under "normal" usage. e.g. 0.800v for example. One way to find out is to Control-L on a point in the curve, then watch GPU-Z for PL green flags when playing.

Once you find that point, then flatten the curve completely at that point all the way to the right.
Now this will, IIRC, cause both PWR and Voltage flags to be constantly lit on, but the GPU clock won't fluctuate anymore, because it's not at the "Real" power limit (TDP).

I tried this at +183 mhz offset, and 1.025v @ 195W, and it triggered both PWR and VREL even though I was far below 195W. If this is similar to what Bennyg is saying, then that's all about manipulating curves (if this same trick works at 0.800v).

 

This make no sense, with a 1060 I have much more / to much load (and temperature) at the identical graphic-settings.

@Robbo99999 & @Falkentyne
Thanks for your opinion, but I don´t understand all. Also: The tool in this thread = you mean the hardware (dissamble the notebook) to read and make a custom bios?

My actuell setup, maximum constant original boost (1733 MHz) stable at fix 0,8v (make with Afterburner Curve Editor) is very better than Original. Because cooler and constant and high clock. 1800 MHz for example doesn`t work.
But: I think 0,75v for 1733 MHz would be stable, too. Or 0,65v at 1550 MHz. This all make sense, yes.

The battery-mode / behaviour would cost too much power. And: the less voltage (0,65-0,75v) would be stable with more and constant clock. The battery-mode für ac-mode isn`t a good idea, yes.

I think I have only two good possibilities:
1) be happy with my actual and cooler setup (1733 MHz and 0,8v) without dissamble my notebook
or
2) order the hardware to make a custom bios to find the most optimal settings.

 

This should be my actual setup.

Ok, 1733 MHz is not the highest possible clock for 0,8v, I think 1750 for example should be ok, too. But 1800 MHz is to much (tested), for example.
But a few MHz should not matter... 1733 MHz constant at fix 0,8v is the highest possible clock at Standard-Configuration (15XX-1733 MHz with high 0,8-0,91 voltage...)

 

Interesting, although I never experienced stutters or any lack of smoothness when experimenting with lowering the Power Limit of my desktop card - and it was bouncing off the limit.

Yep, by the "tool" I meant the software in this thread as well as the disassembly of the notebook and flashing of the vBIOS using a hardware programmer. I suggested doing that to lower your power limit and overclocking at the same time - that's how I get most performance per Watt out of my desktop card when I was playing about with it (...I didn't need to flash a vBIOS, with desktop cards you can just adjust the Power Limit in software like MSI Afterburner). I'm not a fan of the ctrl-f voltage curve editor in MSI Afterburner, I've read about it, and tried it, and it does nothing for me apart from producing some very strange results when things get out of whack, but there are a couple of people who find use in it. If you can be bothered to flash a modified vBIOS with a lower power target then I think this is the way to go for you - but only if you want to go to the exrtra work of it all, then combine that lower power target with overclocking to produce maximum efficiency per Watt.

EDIT: I don't have experience using this software 'tool' in this thread, nor with the hardware programmer, so others can advise you on that, but in this post I'm just talking the theory of the approach, in terms of what I would do.

EDIT #2: you might also want to try to improve your cooling before you do any flashing, you might get lower temperatures by repasting your laptop or using a laptop cooler, then you might not need to lower the power limit.

( @Robert Kattner , I edited my post after you replied to it, just to bring your attention to the other points)

 

General the Curve Editor from Afterburner would be good for me, but the smallest configurable voltage is 0,8v....;(

Now I must order the hardware programmer and use the tool, because I don`t like to give away potential. ^^

Thanks.

 

Does anyone know, which hardware programmer I need for the Alienware 15 r3 with 7820HK and 1080-MAX-Q?

@leeloyd
Can you help me? I don`t find (anymore) the right post from you with the hardware-programmer-infos.

Thanks.

 

Already done...

- Thermal Grizzly Conductonaut for CPU
- Thermal Grizzly Kryonaut for GPU
- undervolt cpu (-100mv)
- custom-base with (silent) active fans
- and 0,8v maximum voltage for 1080-MAX-Q

Result:
- under 80 degrees at 99% (nonsensical) Load at silent 2500 RPM at fans

But I play my games with 65-85% load in average (reserve for heavy scenes). The temperatures are usually 60-70 degrees (or 70-80 degrees with very silent 2000 RPM).

Another advantage by bei playing games at 65-85% load: never get a power limit. 0,8v maximum with curve editor is often 0,04-0,1v fewer. (without reaching power limit the 1080-MAX-Q-voltage is often 0,8X-0,91v).

0,8v together with a wise load (65-85%) is clear cool / cooler than:
- 0,7-0,8v with 99% load or
- 0,8X - 0,91v (standard without reaching power limit) with 65-85% load

But still I will (more) undervolt my 1080-MAX-Q to 0,65-0,75v, because there is still potential. Yes, it wouldnt have to be. But I want to.

 

What are you talking about, a 1060 has way less heat output than a 1080. Why else do you think there are 1080s in large notebooks and not in tiny ones with crap cooling? In terms of efficiency the 1060 is king. There is no card more efficient than a 1060.

 

@Danishblunt
However, let's not talk about it anymore. Anyway, I do not want to sell my notebook with 1080-MAX-Q and buy a 1060.
Furhermore I have the choice with a 1080-MAX-Q: louder at high(er) perfomance, or silent with more perfomance (but lesser than the louder-configuration) than 1060, too.

 

And laptops with 1060 run much colder? I who thought ODM / OEM*s designed the cooling around the hardware used

 

Yep. Various people have used different chip flashers but CH341A (black pcb) and SOP8 150mil adapter cable that cost all of 15 bux all up works for me.

If you power-limit the card, a + core offset forces the card to choose a lower voltage for the power-limited maximum clockspeed. That may take a fair bit of playing around to find full stability since you're dealing in the neverland between idle voltage and 800mV where you have no curve to see, let alone edit

 

Thank you. I have to think again, I have no desire to open the notebook again. And the advantage is low (in relation to the effort), also because I could do optimizations, repasting, max 0,8v with curve editor etc. (see #961 for details), the notebook is already very quiet (and cool enough, usually around 70 degrees).
The custom bios would make the chip again a few degrees cooler, but I would not reduce the fan-rpm anyway. I don`t have as much benefit as the optimizations I've done.
Let's see. If I do it, I will post here again.
Thanks to all!

 

it's got nothing to do with the laptop manufacturer and all to do with what chip manufacturer and model is being used for the VBIOS EEPROM.

 

I did some brief testing on my desktop GTX 1070 yesterday with a 50% TDP Power Limit combined with my max stable overclock. I ran a power hungry GPU benchmark - Superposition 4K, and the card was constantly bumping against the power limit with the lowest reported voltage during the actual run being 0.781V at around 1600 - 1750Mhz, most of the time the card was running at 0.8V'ish. So, yes, combining a low power limit with a high overclock will force your card to use the lower voltages, which is more power efficient. (My GPU fans didn't even turn themselves on during that benchmark, with a 50% Power Limit the max power consumption would be 110W for my card).

EDIT: Just now, to prove the concept of lowered voltage even more when using this method, I ran Furmark with a 50% power limit combined with max stable overclock. What I saw here was frequencies from 1265Mhz to 1330Mhz at voltages from 0.662 - 0.675V; so using this method of combining a reduced Power Limit with a max overclock will allow the GPU to use ultra low voltages if it needs to. (Idle voltage is 0.650V on my card, so it was running only slightly above that for the Furmark run).

 

I used a Mini Pro TL866A with 1.8V adapter and SOIC8 5250 Pomona Clip.
Eager to know how your 1080 Max-Q will be handled by @Coolane 's Pascal TDP Tweaker

Edit : Did you try to Dump its VBios with NVflash and to open it with PTT ?

 

No, but now, and it works, see attachment.
And now? Can I modfiy and flash the modfied rom with nvflash without hardware? I don`t think so (?)

 

you can mod it, but you can't flash it on the card without hard programmer.

also you can dump BIOS via GPU-Z as well.

EDIT: it's quite weird that you don't have any numbers but 0s. Maybe it did a bad dump or something.

 

I thought so. Ok, i stay at my last opinion.

No, that didn`t work.

Yes, something isn`t right, I think so too.

But ok, I stay with my current setup

 

@Coolane could be interested in your GTX 1080 max-Q VBios to improve its software.

 

If he has the subvendor etc missing after flashing a modded vbios its because of the IFR header is missing as the Tool coolane made removes it to stop the blank screen issue.

I suffer the same issue on my MSI gtx1060 once nodded it looses the 1462.11FF part of the device ID however the rest is normal.

Sent from my SHIELD Tablet K1 using Tapatalk

 

odd, my MSI 1060 lets itself get dumped without any issue. All information there.

 

My gtx1060 let me dump but modded firmware removed vendor ids. If nvflash can't dump firmware then there is an issue with any changes made to a max q vbios.

Sent from my SHIELD Tablet K1 using Tapatalk

 

nvflash can dump, just not flash modded firmware. there must be a flag or something that's specific to max-q that gets removed at some point

 

if it nvflash has dumped the vbios but they can not flash it is the user sure they are connected to the vbios SPI chip? Guessing that Max-Q is BGA'd to the mainboard here.

If they are connected to the correct SPI chip then they have implemented a change on how the vbios is stored which is bad news, Some if not most new SPI chips have two area's one is FLASH the other is OTP. OTP is write once and can be used for anything including protecting the SPI area from programming.

However the user must first check that the chip he/she is connected to is the correct one with 3 reads and saves (Flash and OTP area) to compare with to be on the safe side. Then try to flash if its the right chip. If it doesn't work then check connections and try again, if it still fails then Nvidia has changed something and your stuck with what you got.

 

Someone on this thread previously mentioned locking the clock frequency to a specific point in the curve using MSI afterburner (I believe it was @Falkentyne ) so that you get more consistent frame times and frame rates and to lock it to a point where it is below the power limit you have set for the card.

Can I just say how MUCH this has improved game-play performance of my card, especially after unlocking my 1070 to full desktop power class.

Previously I kept bumping up to the power limit and constantly noticed hitches and bumps in game-play but was not sure where they were coming from. But ever since locking the frequency and making sure it doesn't bump up to the power limit, game-play has changed NIGHT and DAY. Now there are no more hitches or bumps and everything runs buttery smooth.

Clocks are stable and not jumping up and down anymore. To anyone noticing stutters and hitches during game-play, try out this sweet fix. Thanks @Falkentyne .

Now am locked at 1911mhz @0.9620v desktop grade performance in a 15inch notebook

 

@Timbabs123
That was me who said that

 

Oh my bad edited

Thanks mate

 

Glad it worked!

 

MSI GTX1060 120W just crashed
the default BIOS MSI is 78 W

 

Running my 1060 now on 120w it sure fly's when on stock clocks.

Sent from my MotoG3 using Tapatalk

 

hacktrix2006
I would advise to put 100W for non-extreme people))))))
my settings are GPU +175 MEM +500 TDP120W (accurate TDP still in search)
120W just crashed

 

I have mine set to 84w for normal use and set to 120w when i go for benchmark runs with OC's, Its not the first time this card had felt the sting of Overclocking certainly wont be the last.
Whats your temps like on the +175/+500 offset?

 

auto fan 80-85 degrees P870DMG
110W so far so good ....
tests I went to 140W)))))))

 

Back down on your core overclock. Maybe you are unstable.
At higher TDP, card will boost to a higher clock. Possible it will pass a "voltage point" where that higher clock is not stable. if this crashes the video driver or game, that is the reason.
I am not sure, but I think this can also happen if you reach a new PWR limit, and the voltage drops below power limit, but the clocks do not drop, then the clocks are too high=crash.

If you get black screen crash, it's VRM current exceeded.

 

Falkentyne
fankyou )
The problem in power consumption in overclocking now I'm playing on TDP 130W is holding but overclocking GPU+0 MEM+500
you need to pick up the acceleration under TDP

Sorry my English ((((

 

Raising the TDP doesn't mean you have to get rid of your core overclock completely, maybe experiment with smaller increases in core clock. Before you were at +175Mhz core overclock, maybe try +100Mhz combined with the 130W TDP and see if it's stable - if it's not stable lower it to +75Mhz and try again, then if crashes try +50Mhz...you get my point. (Hell, maybe even +150Mhz is stable, you don't know until you try it).

 

Yes @Falkentyne has it.

The higher power limit may be exposing higher boost clock/volt points that aren't stable at +175, but are never reached under load at a lower power limit.

Just dial back the offset a bit until stable.

The way the offset works, to be completely stable at a given offset, every clock/volt point has to be stable.

Depending on how you test, you can achieve a false positive of stability by not testing at the variable loads a game may put on the card; for example kombustor on my 1070s settles at 1100-1200MHz at stock TDP, but its pointless to test here since this is an artificially heavy load NEVER seen in a game, where clocks settle at much higher on the boost curve, 1700mhz minimum. I use Firestrike stress test, 1600-1700mhz at same power limit, but its a bit light on CPU compared with modern games.

 

Robbo99999 bennyg
fankyou for advice ....
for myself I decided to first find out the stable frequency and then raise it TDP
TDP 110W GPU +200 Mem +500 while stable))))

 

@qwerty8224 You might want to try the following settings +100 offset on Core and +300 offset on Mem, Just done this on mine and benched it, https://www.3dmark.com/fs/14822817
Going to continue with the OC's on the card to see where the limit is on GTX 1060N's for MSI MXM Version. To me a 175mhz bump on Core and 500mhz on Mem seems slightly high to be even close to stable.

 

hacktrix2006
TDP 120W ... a lot for the 100W cooling system p870DM t~90C cooler max (gelid extreme)
witcher 3 stable gpu 200 mem 500 tdp 110w temp 83C

 

The vBIOS on my card has setup like :



So normal use is always 85w but using the slider i can go to 120w (141% on MSI AF Slider for power) also have the Thermal changed as well so it shouldn't have issues if getting hot. I also use a Custom Fan curve which keeps even 120w lower then 85c. I do however agree that the HSF is not upto 120W which is why i only hit 120w TDP when benchmarking in bursts after all my HSF is infact a 75w TDP one as its the same as the GTX 970m HSF. Thanks for your warning though will keep that in mind.

I am using Kryonaught and Minus Pads for GPU and CPU 0.5mm / 1.0mm and 3.0mm with the GPU to CPU Bridge using 2.0mm on the CPU HSF connection point.

 

Msi gt75vr , unfortunately doesn`t work . Changes of tdp for gtx 1070 doesn`t give performance boost .

 

Ok my fault in bios it should be also PEG TDP higher than default 75 ;/. Now it is working .

 

@majster msi why do you have to change TDP SLOT POWER in the Bios? That doesn't do anything....

 

it does . Do you need proof ? Gt75 has some improvements in bios.

 

Yes, please
Thank you.