Mobile Pascal TDP Tweaker Update and Feedback Thread

SYSTEM AGENT->PEG PORT bios options are identical in GT75 and GT73....there is nothing different.
What did you change slot power to?

I changed slot power from 75W to 85W and got a 22,100 graphic score in 3dmark.
I'll change it back to 75W and try again

*Edit* changed back to 75W.
Same score (22,148 graphics score).

*EDIT*
WHY ARE YOU USING FURMARK?
If you want to test TDP, use 3dmark, firestrike (normal) and look for graphics score.
Furmark doesn't tell you anything.

And make sure Nvidia is using default graphic settings. 16X anisotropic filtering in the driver causes 3dmark invalid score now. Must have been a new patch, that didn't happen 2 months ago.

 

New height reached,

Got 2000mhz Stable with my BGA 1070 D. This is just fantastic for a 15inch laptop especially considering the size (desktop performance in such a tiny chassis)

Locked at 1.043v on the curve and consumes about 180w (its a bit of a gozzler) and max temp is about 85C

I had to downclock my CPU from its 3.8ghz with -130mv to 3.5ghz with -145 to allow more power budget to the GPU to get it stable but I don't mind as am mostly GPU bound at the res I use(1440p 120hz), and its just 300mhz so minor.

I also unlocked addtional power budget to get it to the 2ghz , so up from the default 150w (target) - 175w (additional) to 150w - 195w.

 

Cool, what's your Firestrike Graphics Score now? Good decision to transfer the power budget from CPU to GPU.

 

Fire-strike: 15399 up from my previous highest of 15187, link: https://www.3dmark.com/fs/14865459

Time Spy: 6169 up from my previous highest of 6087 (actual highest I got was 6177 but for some reason that didn't validate): https://www.3dmark.com/spy/3260060

I believe it can do more but I just have so many background apps going on which I cba to close.

*EDIT*
I rank 7th (top 10 wow) now for highest time spy with a GTX 1070 and highest Clevo overall in both time-spy and fire-strike

And I think I can get higher but not sure if its just my background apps and windows saturation that's holding it back.

 

Good scores for laptop GTX 1070, but was expecting about the same score as my overclocked desktop GTX 1070 due to your greater number of cores - so thought you would get 21000 graphics score rather than 19400 if you're locked at 2000Mhz on the core. It didn't drop core clock at all during the run did it? You also have pretty much same VRAM overclock as me too, so it's not that. (21,359 is my best Firestrike Graphics Score, thought you'd be closer to that). Good scores though, and your Timespy Graphics Score is the same as mine!

 

I think the reason why my scores aren't as high is because maybe my CPU isn't clocked as high or I have too much **** running in the background and no clocks don't drop though they do change.

Generally clocks switch from 1987~2012~2025~2000 during the run.

And yeah I did expect it to perform a bit better, I might do a few more digging around and see what I can find plus any suggestions are welcome!

At least its great my Time spy doesn't disappoint too much.

 

Graphics score have nothing to do with CPU. You can only blame low combined score on CPU.

Likely power supply or EC.

 

Timespy score is fantastic, equal to my max stable overclock on my desktop GTX 1070 which is a Zotac AMP Edition card. Try setting "Prefer Maximum Performance" in the NVidia Control Panel for 3DMark - I gained nearly 5% by doing that (but only in Firestrike).

 

Alright thanks I will try do that and give it another run when I get home, currently at uni.

And shweeet, that's impressive for a mobile chip

Na I doubt it EC is not that much of a problem now, I monitored the frequencies when it was running as well and CPU didn't really drop from its 3.7ghz.

 

I got 22,000 graphics score with a 1070 @ +183/+700, 195W TDP, voltage locked at 1.025v, or if i do NOT lock at 1.025v, then instead curve flattened manually past the 1.025v (then I can downclock at idle).

You should be able to run +150/+500 easily like this, lock at 1.025v, flatten past 1.025v. You should be getting 21000+ graphics score.
Make sure your Nvidia driver settings are default, and you can try the older 388.00 driver, or the newest one. One recent driver caused low 3dmark scores..

 

Where does it stands @Timbabs123 talk about Graphics score? From what I can see in those two posts, he complain about the TOTAL score...

 

Good scores!

You can see the Graphics Scores in the links he posted - we've been talking about the graphics score. We know his total scores are lower because of the underclocked CPU, but we've been talking about Graphics Score.

 

do you have a link for that?

 

link for what? my last score? I didn't save it. I'll do another run shortly.

 

yes, a link with that score

 

ok.
@Papusan

https://www.3dmark.com/3dm/25024053?

 

Oh’well, Physics is a bit low for 47x

 

Good scores! Essentially identical to my desktop, for you a tiny bit more on the GPU and a tiny bit less on the CPU.

I get 15200 odd Physics score at 4.7Ghz, and that's with faster RAM. So, I think that's down to the RAM. Kaby Lake doesn't have higher IPC than Skylake does it?

 

Nope. Equal IPC. But 2400mhz running low timings and 1T should manage higher physics.

 

Errm, yes! I noticed gains in Physics score by tightening my timings on my already fast (in comparison to notebook) RAM - so the difference between my CL14 3200Mhz RAM and fairly unoptimised laptop 2400Mhz RAM is going to be a pretty big difference. I think this is a large part for my slightly increased Physics score over Falkentyne's equally clocked Kaby Lake CPU. So I think his CPU score is just fine for his 4.7Ghz clock rate.

 

I haven't speedy ram, 2666MHz and 15185-15200 with 4.7GHz in FS. He run 2400 with faster T1. Shouldn't make the very big difference in FireStrike (3DM11 is another matter). I don't say he should manage 15200, but barely over 15050 with slightly slower ram speed is too low. But that my meaning about it

 

Well, whichever way we look at it it's only a 1% difference between 15000 and 15200, so he shouldn't lose any sleep about it.

EDIT: OK, @Papusan and @Falkentyne I had an itch to satisfy my curiosity, so I ran 2400Mhz CL16 vs 3200Mhz CL14 RAM on my desktop right now. Here are the results, the 3200Mhz RAM provided 1.3% increase in Physics score, jumping the score from 15000 up to 15190. So, I was right in thinking that fast RAM can make this small but significant difference - if I set my RAM to same speed as Falkentyne's then I get the same Physics score as him - his CPU score is fine for his clockspeed.
2400Mhz CL16:
http://www.3dmark.com/fs/14870976
http://www.3dmark.com/fs/14870988
3200Mhz CL14:
http://www.3dmark.com/fs/14871037
http://www.3dmark.com/fs/14871050

 

Then you should have higher Physics score than what I get from 4.7GHz https://www.3dmark.com/fs/13044865. With your explaination, your faster 3200MHz CL14 ram should kill my Physics score done with wimpy 2666MHz and CL15+2T

 

what TDP?

 

Ha, knew you were gonna say that! ;-) Maybe you've killed off some Windows 10 processes and running no Anti virus. But, I'm not gonna lose sleep over it, and RAM does make a difference, I proved that.

 

@Papusan
I changed my tREFI to 65535 (maximum), it was at 32767 before, ran memtest86 for an hour to make sure my OS wouldn't go byebye,
then,

I got physics score of 15,131 on the very first run after rebooting windows. But then afterwards it was close to around 15,062 every time.

I don't get it....

cool. after posting this, closing browser and running fs again I got 15,018.
Ok I'm done. You guys have fun.

 

HeHe. Who would run benchmarks with Anti virus running?

Have you tried up the clocks instead of 2400MHz and T1 ? Try lowest timings and T2 with 2666MHz. I think you have tried this before, but not tested this in 3DMARK. Correct me if I have wrong.

 

2666 mhz won't POST unless I set timings to 18-19-19-43 2T
and that makes my latency go through the roof. Up to 57ns instead of 50-51ns.
My write speed goes up to 43000 but this isn't worth it.

 

Yeah, I agree, not worth it if it increases that latency. I heard that low latency is the bees knees when it comes to gaming!

 

It's a combo, speed and latency combined. Divide the speed by the latency to get the relative latency for comparison.

 

The point was to see if it made a change in FS physics. Not saying you should run it permanently. And I can't tell what timings you could encounter with the change. Have no crystal ball

 

Ok I'll try it later. I hate boot loops
@Papusan
I just spent $684 on a 1 TB Samsung 960 pro SSD and put my steam folder on it. Now I have to make my jokeBGAbook boot loop and crash

 

Finally, can I say something nice regarding your notebook jokebook. It contain a damn nice ssd inside, but expencive. Now You will have a very nice drive to put in your next laptop

 

Yep, when I was talking about latency I was thinking along the lines of the actual measured latency part of the test in Aida64 benchmark, which I think takes into consideration what you're saying there (e.g. 41.0ns in my case in below pic):

 

Inb4 I try CL 14/15/15/35 1T tREFI 32767 tRFC 270 and wind up reinstalling windows after the attempt....
@Papusan @Robbo99999
You KNOW it will either: 1) fail to POST, 2) trash the OS completely if it does POST....

 

Before you try it, maybe back up you OS partition using Macrium Reflect which is a free back up program. I have my OS and programs (not games) on a seperate 60GB partition, and then all I do is use Macrium Reflect to image that partition to an external USB HDD drive - that way if anything goes south from tweaking attempts you can just easily roll back to a stable non-corrupted operating system without having to go through the drama of a whole new Windows installation.

If you're not gonna do that backup, then I'd suggest tweaking your RAM somewhat gradually and then checking with Memtest86+ on a bootable USB stick so that you never even enter the windows environment for initial stability testing of a new RAM overclock - keeps OS safe from corruption. I'd recommend doing the stuff in this paragraph regardless of backing up with Macrium Reflect or not. After testing with Memtest86+ I recommend testing RAM stability with a program called HCIMemtest ( http://hcidesign.com/memtest/), using a seperate instance of that program for each thread of your CPU (so 8 instances for my CPU for instance - i7-6700K), while adding a figure of RAM amount to be tested into each one so that it equals about 90% of your total RAM being used. (essentially following the instructions in this DDR4 overclocking guide: http://www.overclock.net/forum/5-in...-intel-ddr4-24-7-memory-stability-thread.html)

 

I'll do that now. I did run memtest86 yesterday for testing tREFI 65535, and got 15,131 physics score on the first run then it dropped back down to @Papusan 's lovely 15050 which he hates....so no point in risking RAM corruption if it gets hot (tREFI is how much time the RAM can spend without being 'refreshed', while tRFC is how much time the RAM can spend doing "nothing", so lower tRFC is good and higher tREFI is good, if the RAM doesn't degrade data from not being refreshed fast enough.

Ok it actually posted at 14-15-15-35 1T tREFI 32767 tRFC 270 @ 2400 mhz
And this is kingston valueram ....

You know this isn't going to be stable.
Memtest is loading now. Waiting for errors.

 

@Papusan
I tried 14/15/15/35 1T, tREFI 32767 tREF 270 2400 mhz

Write (RAM) speed seemed slightly increased, but no change in latency. Physics didn't change. Probably other timings causing IMC adjustment because 14-15-15 timings don't add up to TRP ratio (this is supposed to be: CAS + TRCD + TRP = TRAS). Not going to bother. Last time i tried messing with this stuff had trouble posting. Although previously i had TRP=5 and TWR=10 (TWR is supposed to be double tRP or NO POST).

Then I set 2666 mhz, 18-19-19-2T. Physics slightly lower, like 15,030. Tried again and 14,990. Not worth.

 

It's ok to get things confirmed for your own sake. It is a lot worse live in ignorance At least you have tried.
4.7GHz and 2666MHz CL15 https://www.3dmark.com/fs/14876435

 

I tried with the same flasher without success tho. It can't seem to read the Bios chip, neither can I find it in the program, I've downloaded from the youtube video like you. What am I missing here?

 

Use the programmer without the 1.8v adapter BUT only do this if you are using that exact model. Also make sure it's connected properly to the bios chip.

Use the latest version of the software and make sure you get the correct name of your bios chip, then select that name in the software.

Sent from my SM-G935F using Tapatalk

 

Do you have a link of the latest software somewhere? I downloaded the one from the youtube video and it seems old.

 

I will have a look at my old PC for you later on. See if I can upload it.

 

Anyone is able to edit a vbios to put a max vddc on 0.8750V? I dont need others things such tdp, i just need that GTX 1070 not go high than 0.8750V on vddc.
Vbios is 86.04.5b.00.44

Thanks.

 

Don't need vbios mod to do that.

In MSI Afterburner:


Note, dropping the clock on the next higher voltage lower will snap to the same clock and give you the flat line, most of the time. It saves a lot of time. That GUI is atrocious

 

​

Yes, but my objective is not use a external program to lock the vddc, anyways there is a way to edit a pascal bios to edit the max voltage and flash the mod vbios without any hardware piece?

 

Hi!
I assume flashing a modded vBIOS is still not possible with software.
According to NVFlash, the chip used for my vBIOS is a WBond W25Q80EQ 1.65-1.95V chip.

Also, the vBIOS I am trying to flash is for a full BGA ASUS G752VM laptop with GTX 1060.

When I load the vBIOS I get this:



However when I open an MXM vBIOS, I get this:


There are 3 unknowns for the BGA ASUS card...is it safe to mod it like that and flash it? I am afraid that the software is bugged and it would not save a working vBIOS.

ps: also tried it with my desktop GTX 1080 and that is detected correctly as well. The software seems to only have a problem with BGA vBIOS.
Is it safe to use it despite this ?
I am not even sure that the voltage numbers are deteted correctly ... oh and preset button does not do anything

 

@Coolane ?
I think the tool needs to be updated for other card revisions. Although that's not easy without access to hardware.

Maybe if you send him the Bios dump?
But I haven't seen coolane around recently.

 

i do have a question if anyone can help me out. i have a gt73vr msi laptop. i have the 1080 card. would it be possible to change or increase the temperature to a higher setting for Throttling. i feel that these cards start throttling at such lo temperatures.

 

I already answered your question in the other thread, yet you didn't even reply to me or acknowledge my post. was it because of that troll Danishblunt? You realize he doesn't know what he's talking about right? He's the guy who claimed he made his own custom EC to make a MSI GT73VR run STABLE at 5.2 ghz at 1.270v on TRADITIONAL Thermal paste (which is impossible).

Yes, you CAN Change the TEMPERATURE (critical) throttling point. You can raise that up to 80C or something. This throttling is shown as "limit reasons: temperature limit", as far as I know.
You can NOT change the "Pascal" throttling "Feature" which throttles the core by -13C every 8C temp step, starting at 42C. This is by design and does not show up as a throttling flag, and there is no way to avoid this, on either desktop or MXM cards.

If you actually want to TDP mod the 1080, which no one has done on a GT73VR or GT75 yet, I suggest pressing the 1080 'preset' button in the editor (otherwise your max TDP will be 200, regardless of what you enter in the editor), and then try a range of 200,000 to 240,000. No one here can tell you if going above 200,000 is safe on a GT73 or GT75, however. Clevos have been tested safe by Khenglish.