Mobile Pascal TDP Tweaker Update and Feedback Thread

Ok I have a serious question for you guys.

Does anyone here know if the laptops identify which video card is inserted (via EC firmware) via the "device ID" like 1462-10B1 or 1462-17FF, or via one of these codes strings I found sitting in a few Vbioses:

750029150010 (GTX 1080).
725029140010 (GTX 1070)
750029140030 (GTX 1060)

Would replacing one of the numbers with another cause the laptops to identify the card (not necessarily windows, just the EC firmware->System Bios) differently?

@Prema
Thank you!
*Edit*
didn't work. Just not experienced enough going to have to wait for Svet I guess...hey I tried my best...but i'm a newbie

 

The subsystem part of the device ID is determined by the chassis, not the card

NB I use the term 'subsystem ID' to avoid the confusion nvflash causes by calling it "a device ID mismatch" because the full hardware ID string as shown in device mangler is usually
PCI\VEN_10DE&DEV_11A0&SUBSYS_71021558&REV_A1
where you have the vendor (10DE=Nvidia), device (11A0=680M), and subsystem IDs (1558=Clevo 7102=P170EM)

You can override a subsystem ID mismatch in nvflash with -6 (despite nvflash calling it a 'device ID mismatch'!!)
You can't override a device ID mismatch (and say flash a 10DE-1BE1 vBIOS to a 10DE-1BA1 card)

TL;DR if you find a stock 1080 GT73VR vBIOS and use nvflash --version to grab the details of it, it should have exactly the same Subsystem ID as your stock 1070 GT73VR vBIOS does. If not - that is definitely something to investigate...

 

At last 21K FF graphic Score, and new personnal record :

http://www.3dmark.com/3dm/21147719

GPU +212MHz + 512Mhz Max Power 165W Max temp 77°C.
CPU @3.8GHz -100mV Max Temp 86°C max power 49W

Room temp 22°C helped a lot.

Happy, before this VBios mod I struglled just behind 20K. Thanks to @Coolane and the first Pascal BGA VBios modder: Dreamonic from Voltground.

 

Good GPU score, same as my desktop GTX 1070! Temps good too.

 

@Coolane @Mr. Fox @Phoenix @Papusan I think I hacked the EC (indirectly)!

Still need to test.
I was able to draw 150W from GPU and 65W from CPU without PL2 throttling.
Testing again to confirm...

Ok it works. Not ideal but it works...CPU is drawing 0.03v more. Cinebench score is still 1,000-1003 at 4.5 ghz (Ring x42). Still drawing power from battery; don't want to risk raising the slope anymore as I don't want to cause damage. This will have to do until I get a proper EC mod from Svet.

(You guys know what I did, right?)

 

Nice score! Glad to see you are enjoying the extra power!
Maybe you can catch up my GPU score one day (3~5C room temp): http://www.3dmark.com/fs/11661315

 

I am happy that you finally got it working!
What magic did you do to remove the CPU throttling?

 

Changed the Imon offset.
There was a post awhile back about someone making his CPU use double the TDP (147W on linpack), and then him making it use half the TDP. He was talking about editing the EC and changing MSRs to do it. So I started looking through the Bios and first tried something like changing PS Current Threshold1, from 80 (20 amps) to 400 (100 amps). Well that didn't work. The laptop shut down (powered off) and rebooted in 5 seconds after I started a 4 thread load at 4.5 ghz. So I looked at Imon and "slope" and saw it had a negative offset possible, then I read this thread again:

http://forum.notebookreview.com/threads/tdp-and-power-limiting-haswell.766743/

He was tricking his CPU to report both double the TDP and half the TDP it was currently using, by modifying the EC MSR's.

From what little I could find, it looked like Imon dealt with TDP reporting, so I set a negative offset of 25 and enabled Imon Scaling support.
This made the CPU report it was using 15-20 less watts at 4.5 ghz, but the VID was slightly higher. So then I ran firestrike extreme *stress test* (edit) with 150W TDP + 1886 mhz and with a 4 thread chess engine running at the same time (8 threads takes too many resources and the GPU gets starved) and finished the loops and there was no power limit exceeded messages.
Not sure what the 'real' load was or if the PSU was exceeding 230W, but this was enough to make the EC think I was within the power limit.
I guess the EC reads what the CPU reports, and the IMON pin on the VR determines what the CPU reports (the CPU cannot know how much power it's using by itself)

Ran it a 2nd time to verify.
It still draws from the battery (ideally it would be nice if it didn't) but that will have to do until I can talk to Svet.

 

You can buy a kill-a-watt and measure the power consumption from the wall.

Sent from my SM-T560NU using Tapatalk

 

Interesting finding and thank you for sharing that!
I never played with IMON offset/scaling before, didn't know it can affect the CPU power reporting. I think this can be a good way to increase CPU power draw for laptop that has restricted EC, maybe it has the similar effect like VCCIN (in ThrottleStop) that affect power reading (report less power draw).
@bloodhawk , maybe you can try this to remove CPU throttling in your R4?

 

This seems a lot (almost the same exactly) as the "Shunt mod" people were doing on desktop Pascal cards, to bypass power limit TDP throttling....

I set a negative Imon offset of 25.576 (25576 entered in Bios, as per the tooltip).
Pretty sure this is a percent value, as this affected load more than idle.

VCCin=VCCinput...sounds almost like it's the same thing in a way...

 

Once the BIOS is unlocked then yeah. I actually snagged a 6820HK + GTX 1080 motherboard for $500 last night, so going to be fiddling with that as well

 

That means you just had regular TDP throttle before (not even reaching the low level EC limiter). If you have a supported CPU with adjustable TDP this method is not required and you could just raise the power limits (1-4), so that this doesn't affect battery operation.
Once I have blown my W230SS battery because I didn't realize there was a power cut.

 

@Prema

My CPU can do 80W TDP without throttling. I've gone over 85 but then I hit Prochot so I can't sustain it.

All I have to do is raise the power limits and raise ICCMax to 480.
It's a 7820HK.

I've stress tested my CPU at 4.5 ghz for 3 hours straight, at 71W power draw, with 4 cores 8 threads (Stockfish chess engine) as it's less stressful than prime95 (I can keep temps under 85C) while Prime would just ........prochot...

The EC for some reason forces PL2 throttling if the maximum power draw of the system exceeds 230 watts, which then locks the CPU down to 45 watts.

At 4.2 ghz, with -100mv vcore offset, with 150W 1070+Firestrike Extreme stresstest on 1070, and stockfish 4 thread chess engine running, I can stay far below 230W, so I never get PL2 throttle from the EC. There is rather slow drain from the battery (Idk what the total draw is).

At 4.5 ghz (needs default vcore) with FSE stresstest and 1070 with stockfish on 4 threads, PL2 throttle kicks in after about <10 minutes. There is rather high drain from the battery (about 1% every 2 minutes!).

Its' very similar to what you saw when you overclocked to 4.8 ghz on the MSI barebones before the power of Premamod fixed that...PL2 throttle kicks in.

That's the problem.
It doesn't happen if I don't exceed 230W total system draw.

*edit*
With the Imon offset, the 4 thread (not 8 as there wouldn't be enough CPU power available to feed firestrike) chess engine at 4.5 ghz uses about 50W instead of 65W, which is enough to prevent PL2 throttle from reaching total AC system draw of 230W.

If I had a 1080 so my MSI board would identify as a 7RF instead of a 7RE, it wouldn't even touch the battery at 230W; it would be able to use the full 330W (about 260W before it starts hitting the battery).

 

Those are two different limits. PL2 limit is a CPU only limit and not the total EC limit.
If you raise CPU PL1, PL2, PL3 & PL4 and Amps enough and the HK chip doesn't have a hard limit (IDK) it should not PL2 throttle on it's own, but then again it's MSI...

 

Sorry Prema, i was editing my post when you were writing yours.

Yes, I already tried raising CPU, PL1, PL2, PL3, PL4 (platform power limits, anywhere from 200,000 to 999,999--if I go higher, the system won't even POST and I have to clear CMOS) and CPU Turbo Power limits 1 and 2 (max allowed is 500 (500W)), ICCmax to 480, but no matter what I do, once it reaches 230 watts power draw, PL2 just kicks in and there's no way to prevent it from happening.

I spent entire nights trying to fix this, losing many nights of sleep. I finally gave up trying last night.

Now that I found out about Imon, I made the CPU report less power usage, which the EC reads (I suppose), so it thinks it's under 230W and didn't turn on PL2 at all (tested 3 times to make sure).
(remember I'm using the 330W PSU that usually comes with the GT73VR 7RF, but I have a 7RE).

This is not ideal because it still drains from the battery pretty hard (EC is probably reading 220W of usage; 50W from CPU (instead of 65W), 150W from videocard, 20W from "System", but at least it's not reaching the limit and throttling. I could go with a bigger offset on IMON, to try to force less drain on the battery, but I don't want to break anything. Everyone said my "BGA trash" soldered CPU is golden, basically (4.5 ghz @ 1.1v stable) so I don't want to push something too far and have magic smoke come out.

If I can get an EC mod from Svet, then that would solve that once and for all (since battery would be unaffected and AC would be able to draw the full 330W like a GT73VR 7RF.

 

On MSI the CPU PL2 throttles when OCed while running something like R15 without even using the GPU at all. That is not the EC throttle but an additional throttle in the BIOS OS code. It would happen either way no matter if you use a 1060, 1070, or 1080. When we removed it on their LGA model the entire system shutdown during a 5.2Ghz OC bench as it crossed the systems hardware capabilities. @Mr. Fox

 

How much work did you have to do to clean up MSI's spahgetti code? It sounds like you (either had a lot of fun or didn't have alot of fun).

I never had an issue at all with this laptop since I bought it (after raising ICCMax). Everything worked as a Jokebook ( @Papusan ) is supposed to work. I bought the SPI flasher and after dealing with clip issues, flashed the 1070, and only then did I run into the PL2 issue. And I only ran into it at 4.5 ghz on default volts (with CPU and GPU both loaded).

I never once throttled at all at 4.5 ghz with the 115W 1070 loaded, with the CPU also loaded.

Only after I flashed to 150W 1070 did i start having these issues at 4.5 ghz (not at 4.2 ghz at -100mv though).

4.2 ghz on -100mv was a lot less power draw and was flawless. But I literally stayed up almost all night for the last *week*, trying to raise/change power settings to get 4.5 ghz with 150W GPU and 70W CPU draw to work without triggering PL2, and kept failing. Only after I tried to find out what the other "VR" settings were and messed with Imon, was I able to make the (EC? Bios?) think it's using less than 230W.

I'm still wondering though, why is it drawing so much power from the battery after just 150W of total usage? it starts off slow then at 200W it draws quite a bit. Yet I'm using the 330W PSU.

I think it's because (as you said before), I have a 1070 installed, and since 1070 models come with a 230W PSU, there is a cap of 230W (so that people won't blow up their PSU's). that was just my own theory. A 1080 model (7RF) comes with a 330W PSU, so it can draw a lot more than 200W.

 

The additional power draw from battery (which was invented to save money and space on the bundled PSU) is regulated by the EC controller and uses different values for each GPU configuration as the bigger GPU comes with a larger capacity PSU.

Don't even ask...only did it for Fox and EVOC.

 

Thanks. That makes sense.
So I'll just rest and enjoy the - offset Imon (25.276).

I think this is something @unclewebb can add to Throttlestop, as you need an unlocked bios to access it.

I think I can call it mission accomplished. I'm sure my CPU can go higher than 4.5ghz (4.5 @ 1.1v is fully stable) but I don't want to see what other surprises are in store for me if I go to 4.6+.

Thanks for your help and participation, @Prema. BTW the picture you posted of your overclock (was it 5.4 ghz?) was just crazy for a laptop!

 

That was a 4702MQ OCed to 4.5Ghz (2.2Ghz stock base) in a 13" W230SS and totally killed my battery during a bench...

 

Oh no not that picture.
You had another one with a project you said you were working on, that was at least a 5.2 (or 5.3)+ ghz overclock (World record for laptop?)

 

We have user like @Johnksss@iBUYPOWER , @Mr. Fox , @bloodhawk , @Papusan, @Meaker@Sager , @D2 Ultima who have Oced their LGA Clevo's up to 5.5Ghz.

http://forum.notebookreview.com/threads/clevo-overclockers-lounge.788975/page-1096#post-10491513

 

@Prema sweet mother of God!! That on a laptop! Wow!!!!!

 

That's such a crazy sweet deal! Congrats first!
If you are lazy of unlocking the BIOS, maybe try out Khenglish's old post( https://www.techinferno.com/index.php?/forums/topic/6438-guide-dell-e6530-cpu-tdpmulti-unlocking/).
It worked well on my Dell laptop and tablet. And I used this method on a factory 870DM1's BIOS previously to access some settings that are not available in the default BIOS menu.

 

Looks like I was wrong. With more testing, apparently the Imon slope is actually getting ignored by the EC even though the CPU is reporting less wattage. I was just getting 'Lucky" with the max power draw. Svet said there's a huge chance of bricking because a 330W mod is highly experimental. If he think it's some chance of success I'll just go through with it. I don't care if I brick my hardware anymore....

*edit* if I'm being dumb someone please tell me.

 

*edit*
Looks like @Prema was 100% right.

My apologies for doubting you because of cancer MSI bioses.

The EC firmware is causing the excessive battery drain due to the high load and card type, but the PL2 throttling is being caused by the Bios. But it's only happening at 4.5 ghz and at high loads. At 150% TDP it usually takes awhile to happen when the power limits are raised, but at 170% TDP it happens all the time if high GPU load and CPU load occurs simultaneously (it won't happen just with CPU only). I then set the mulitiplier to 4.4 ghz, and increased TDP to 170% from 150% and it didn't happen at all with 4 thread chess engine and firestrike extreme stress test. However after about 5 benchmark runs, I did get something weird at 4.4 ghz at close to 230W power draw:

HWinfo said there was a "IA: Thermal Event" flag, although no throttling occured, and under Ring: Max VR Voltage, ICCMax, PL4, was set to Yes (not currently but sometime during the benchmark run). But there was no throttling at all. But at 4.5 ghz, not only does the Ring one get triggered on, but power limit exceeded on cores and PL2 get turned on also after awhile.

*and edit again*.
It DOES happen at 4.4 ghz but takes much longer to happen. In order for it to happen, PL4 must first be "flagged on", and then load continued long enough for PL2 to get turned on.
*edit*
Setting power limit 4 higher than 999,999 results in No POST and CMOS must be cleared by the 45 second power button depress. PL2 can be set at high as 4095,875 (4095.875 watts) although that doesn't really do anything useful.

 

Man, I'm feeling you're gonna fix your hardware till it's broke! Ha, but no, you don't get if you don't try! Maybe try to get a percent chance idea from Svet for bricking it, then roll the dice!

 

It's what overclockers do right? We overclock till we know things work 100% of the time or we break something (right? @Papusan )
Anyway i did some more testing.
It *IS* the 230W power limit causing the throttling. even if it's occuring in the Bios.
I can reproduce it at lower clocks. It just takes.....awhile to happen sometimes. It's the 230W power cap on the AC despite there being a 330W PSU because I have a 1070 instead of a 1080.

It's probably happening because the battery is being hit too hard or for safety reasons.
I'll bet anyone $100 on it.
But if Svet can't or won't help there's nothing that can be done at all except reduce the card's TDP....

*edit* 100% certain now it's the 230W max power cap because I have a 1070 detected. And I'll bet anyone on this forum $100 cash. I was able to reproduce it BELOW 4.4 ghz at 4 threads + firestrike by using prime 4 threads with AVX + FS extreme stresstest

And just got a reply from Svet.
He's done EC mods like this before but older platforms where he had the board in front of him to desolder and retry until success. Does anyone know where the EC firmware chip is on the GT73VR? is it an 8 pin package, same size as the video card vbios chip? He said attempting to mod the power limits from 230W to 330W (1070 to 1080 type cap) would have a very high chance of bricking and then would need to desolder, flash and resolder.

The only person who could do this successfully is @Prema the Bios God..

 

I don't know about the GT73VR, below are the ECs on Clevo P870DM (borrowed from @Mr. Fox 's pictures: http://forum.notebookreview.com/thr...oenix-has-arisen.781814/page-87#post-10134134). Even though you have a programmer, it will be a pain to de-solder and solder the EC chip to flash it. And you will probably need an adapter for the EC chip.

 

Thank you very much! This is helpful. Yeah I definitely cannot touch that thing........haha.....
Guessing MSI MS 17A1 mainboard is similar.

If I could get ahold of the MS 17-A1 mainboard schematic it would be helpful....

 

They historically used an ENE IIRC, Svet would be the man to talk to about that.

 

So I'm sorry if this has been answered somewhere in the 40 pages..... I plan on tearing down an msi ge62vr to read the vbios chip directly with programmer but I've been so busy, I havent had time to do it yet. If someone knows exactly where the chip is, it could potentially save me a bit of hassle searching for it and I would appreciate that. I hope to get around to it by tomorrow. Thanks guys.

 

On MXM cards, the Bios chip is on the back side of the card (card must be removed from the slot) close to the slot pins. There should be a dot on the chip. It's an 8 pin, very easy to see. They're usually winbond or macronix.(might be impossible to read).

Pin 1 has a small circular indentation. With the MXM slot pins face down, pin 1 should be in the top right corner. If there is a white dot in a corner, ignore it, you want to look for the circular indentation.

Refer to the pictures in the first several pages.

 

If you google image search "sop8 150mil" thats what the chip looks like. Surface mount, 4 legs on each side, indentation as mentioned above for pin 1, usually a small glob of coloured paint on it.

 

You guys have no idea who @pbhenry3 is , do you ? xD

I have been trying to find clear pictures of the motherboard with the heat sink, without success
Will hit you up if i find one before you are able to figure it out.

EDIT - Probably this one - http://i.imgur.com/VKcGB3r.jpg

 

Thanks everyone. It doesn't use an MXM slot. I was not looking for a tutorial or anything, I'm quite familiar with what to look for on a board, I don't need help there. I mainly wanted to know how much work I was getting into, if it was on the main side of the motherboard or If I'm going to have to remove the entire board and flip it around. Or if its going to be hiding under the heatsink like on a razer. (hopefully not) lol. If no one knows for sure by tomorrow, I can update this thread and let everyone know.

 

That looks like the board prior to the pascal series I think, could be wrong but I thought the VR series has 3 heat pipes coming off gpu/cpu. I'll take a picture of the board for you tomorrow.

 

Ah, Even on the Pascal version it seems to be in a similar position, just partially under the heatsink - http://i.imgur.com/i4vGSXK.jpg

Actually there might be 2 - http://i.imgur.com/ULW8hCI.jpg

One as a backup like the newer Alienware Laptops.

If you can make a dump of both the chips it will be easy to spot if its the Nvidia vBIOS or not.

 

got it, thanks for that. If one of those are vbios i should most likely be able to tell which one just by seeing the actual model of chip they are using. Looks like heatsink is coming off in this case.

 

Not unless MSI is pulling funky whoopydoodledoo like this -

http://imgur.com/a/f3JiF (GTX 1080)



Or this (GTX 1060) -



NP at all, glad to be of help. I got so annoyed by the BIOS chips location on my AW 17 R4, i relocated the bloody thing using some 30 awg wires and a SOP8 DIP socket.

 

where's the vbios chip for the p650rs-g?

 

http://i.imgur.com/IbpRoGy.jpg

 

Let's watch pbhenry3's video again, , lol:

 

Maybe re-post the video in the Razer threads

 

Update of my dead 1070:
After Khenglish's initial diagnosis, FETs are fine, the 5V rail is shorted (I guess that's why my motherboard failed to turn on). It's been difficult for him to detect what shorted the 5V rail because chasing it from the mxm slot led immediately descending back to the mxm pins, and he couldn't find any other component that actually use 5V.
Do you guys have idea of what components on the mxm board use 5V or the failure of them can cause the 5V rail shorted?

 

Did he check all the traces lines? Or check the board for fractures....

 

The power phases / VRMs / FETs need 5V's to run IIRC. And on dekstop cards there is a 12V to 5V converter that drops the 12V from the motherboard socked to 5V.

 

Not very sure but seems like he have checked all the possible trace lines, considering he spent a weekend on it. It's possible of having micro-fractures, because he said my board was seriously bent (because I forced modded 980DT heat sink on it and didn't align very well the whole time). He reduced half of the bend and tried to re-flow the top 4 memory chips. If the PCB has micro-fractures then I think the card will have not much hope of being fixed, T_T.

I am going to double check with him to see if he has checked all the power/voltage related components. I looked up some mxm design sheet, seems like motherboard uses 5V to toggle the voltage stability check on the mxm board, then send a higher/lower voltage.

 

The gate drive can vary but is commonly 5V yes.