BigKid

My boring 2 Cents:

NVIDIA simply decided to "cover their tracks" with the 374 driver...

Reason OC is disabled for Mobile Maxwell is that the cards throttled about 20% right out of the box in about one third of systems even when the systems where running very cool. Since this is a sue-able defect in markets like Europe (i.e. not as advertised) they had to tackle it before it became a public topic.

So the only way to fix this issue (since they can not expect the casual user to have a mass vBIOS/BIOS flashing) was raising the TDP limits right on driver level.

The problem, by raising the bar for all Mobile Maxwell was that systems that never had this problem in the first place now the upper TDP bar had also risen by about 20%.

Since all their "lab-tested" numbers (which OEMs build their entire system designs on) are based on the old TDP, they decided to simply remove the core and memory OC option and that way chop of the "same" amount from the top. (Since OC isn't an advertised feature it isn't sue-able).

It is debatable if this raise may have caused actual technical problems, and while it certainly does on heavily PSU and/or cooling limited systems, our tests show that it won't for most of them, especially more powerful ones like the Clevos.

This also brings us to the second and more obvious reason why they approved this solution and pushed it into a public driver. Business!

Why would a mainstream user, who can gain 50% through OC right NOW (through one of our Mods), upgrade to a future "GTX980MX" that gives him only 30% (stock) over his current setup?!

And here they thought they could beat two birds with one stone...Well it looks like they forgot about their end-user!

This is what happens if the management first decides to save money during development and then panics and puts the advice of their legal team over that of their public-relations office...and guess who has to deal with it now.

This is about what I'm thinking too.

1) This bug was there for 5 years - so they are either dumb or plain lying

2) The "fix" only affects maxwell (800M and 900M series) so again - they are either dumb or lying

3) Everyone believing the story about the "bug" will also believe that the false stats released for the GTX970 have been a "comunication issue" - ha ha...

4) For mobiles they are releasing a refresh every second generation - this "refresh" is just a factory OCCed version of the previous one - this saves everyone including notebook manufactureres a LOT of time (testing etc). So my believe is that they are actually keeping potential as a reserve for the refresh (I've seen way to good OCC possible on GTX500M, GTX700M and GTX900M)

My guess is they don't want us to use this potential but rather make us buy a new card every generation (maybe the move of MSI to offer GPU upgrade kits for their laptops starting with GT72 was a reaction to this plans)

90c is the Maxwell design target. 3 degrees are added into the equation for the time the system needs from the moment it begins to throttle to effectively keep the card from breaching 90c resulting in the 87c value.

While it would be a simple tweak to allow the user to raise the temp limit further, I don't believe it to be a good idea.

Other devs may think differently, but they seem to forget that TDP also raises with temperature (voltage x temperature = TDP)

I believe unlocking both is the best combination if you plan to kill the hardware:

http://forum.techinferno.com/general-notebook-discussions/8699-hes-dead-jim.html

@ALL I will try to answer your questions here soon, just need to find the time to dig through 3 weeks of pages...

100% agree with that... It's just... I'd still love to see a version for us GT72 owners with a PT of just about +20% (vs stock bios) and in return a little higher temp limit. As we are facing the EC firmware starting to throttle... So I'd like to test a vBios that tries to stay within (as much as possible extended) PT limits (PT in watts not temp). I know that it's against your conviction but I'd still love to give it a try.

I think having the GPU throttle now and then to stay within it's PT (power draw/watt) is still better than to have the EC throtteling because the impact seems to be much worse... Hitting the temp target also seems to be worse because when it's too hot it will take some time until it cools down again... The PT (Watt) seems to react more quickly and help to not overheat in the first place...

I also do not believe that 90 degree are more dangerous to the GT72/GTX980M as 92 and even more degree where perfectly fine for the GTX780M and GTX880M ... So with the PT beeing active (but a little extended) again I think it would be safe to upp the allowed temps a little in return... Again that's for "us" GT72 owners having the EC firmware throtteling issue...

Prema made the mod to start throttling at 87C by design to have a safe thermal limit but that sucks about the rest. I thought the GT72 was supposed to run super cool?

Sent from my Nexus 6 using Tapatalk

Yeah... It was supposed to run cooler but actually...

With the older Model we had the thermal bridge between CPU heatsink und GPU heatsink so the heat was transfered both ways and got transfered to both "cooling sinks" besides the fan - but we only had one fan.

With the GT72 we have the thermal bridge from CPU to GPU "cooling sink" (not GPU heatsink) - but we have two fans now and the "cooling sinks" are more separated. So heat is transfered from CPU to GPU but not the other way as it was with the older model. The fans are controled individually and the CPU fan is only speeding up when the CPU is getting hot.

In situations where the CPU is adding a lot of heat the new solution might be more powerful.

But in standard situations it seems like it isn't really much (or even less).

The lacking of optimus contributes to this so in "office use" the GT72 is even louder than the older models. The GPU fan does never spin down which is strange considering that the older models have only one fan to cool CPU and GPU and DID spin down.

But that's about the 87 degree beeing Premas choice is contradicting something that I've been told (or understood - so I might be wrong). What I learned is that the 87 is a choice of Nvidia that can't easily be circumvented.

*sorry for going OT -I will leave it here*

Final Update for the other MSI GT72 owners out there:

I bought and connected 300Watt PSU from Dell and while it works perfectly fine the GT72 doesn't care about the +70 Watt available and still switches to hybrid mode.

So I've seen several reasons for throtteling with the GT72 now:

- GPU thermal (when reaching 87 degree celcius)

- GPU PowerTarget (when using stock bios or bios with PT enabled)

- Total power draw - AC adapter shutting off to protect itself

- Total power draw - EC controler throtteling the system

All but the second are also happening with Premas vbios - which is not Prema to blame ! He is doing great work with his mods... Just stating it...

Side Effects:

- the system seems to switch to hybrid mode at a certain point which is well below 230 watt of the default AC adapter - this means that the battery will be used to to compensate power exceeding this (unknown) limit (I estimate it beeing at somewhere around 200 Watt)

so is there mcuh difference after manually appliying thermal paste ? and which kind of thr.pst did u use ? ceramic ,silk etc..

It's borderline... With the new paste job the temps are about 3-4 degree lower which is just enough so it doesn't start to throttle :-)

I' using either arctic silver or notua nt-h1.

sounds strange, in my clevo with 3920xm with 4.2ghz and a little ov and with 980m +135mv ov i dont experiance problems with my 220w psu...

but i get crashes due to very high oc....mybe it s the psu that cant deliver enough power...which wuóuld be nice as i can oc further with a new psu but i guess its very unlikely as those problem just appear with very high oc

Hm ? What's strange about that ? More OC -> more power required... More speed, more heat (even at the same voltages, higher frequencies require more power...). So IF you AC adapter is too weak it will be more prominent with higher OC.

MSI GT72 with GTX980M and OC:

Did another few runs and here is what I found out so far... I posted a similar post in the clevo subforum but I feel it its better to post that here...

With Prema's bios and SVL7 bios at stock clocks

- the total power readings of the "MSI dragon gaming center" seem to be quiet accurate (I compared them to readings of my power meter)

- the power consumption with premas bios is exceeding 245 Watt

- with the stock ac adapter the GT72 is constantly switching to "hybrid" power - meaning that the battery is used in addition to the AC adapter to compensate the high power demand (this can also be seen in "dragon gaming center")

With SVL7s bios at stock clocks:

- same story - in addition I even needed to throttle the power limit slider down to 50-65% - otherwise the GT72 throttled

With OC (both bios mods)

- before I redid the paste job I also saw a lot of throttling when using any OC due to overheat (exceeding 87 °C)

- with SVL7's bios I needed to up the voltage slider higher than with premas mod to be able to OC (also I assume that this is due to different base lines - I didn't compare the actual resulting voltages)

- overall OC seems to be possible with both but there is much more trial and error with SVL7s bios and GT72 owners seem to need to lower the PT slider A LOT - lowering the temp target does help too but gives much worse overall scores / results

- the power consumption in dragon center starts to show "AC" instead of any actual value during test runs - this might indicate that it is exceeding all limits

- hybrid mode is active permanently throughout all tests of 3dMark Firestrike

Stock Bios - Stock Clocks:

Did a few test runs with the stock bios and it seems not to exceed 210 Watt whereas the OC bios reached 245Watt (and probable more depending on what "AC" is meaning in dragon center).

In respect of hybrid power I saw it only triggering once for a short time during my 3D mark runs.

Stock Bios - Max OC (which isn't that much +135Mhz GPU, +600Mhz Mem, +12.5mV).

Hybrid power triggered more often and also for longer periods, yet still less often as with the OC mods (even if those run at stock clocks). 220 was the highest Wattage I've seen.

Question is - why does Hybrid Power trigger at all - isn't this supposed to be a 230 Watt ac adapter (19.5Vx11.8A) ?

Bottom line for MSI GT72:

- a stronger AC adapter might be needed - the GT72 seems to have "dell compatible" connector - I sucessfully tested a 240Watt ac adapter from dell - but it's actually shutting down a little earlier than 240 Watt (might be because its old) to protect itself compared to the MSI one , question is if the GT72 will detect the stronger power source and act accordingly

- with the GT72 switching constantly to hybrid mode and thus also decharging the battery and possibly using up more recharge cycles I'm wondering if OC is a good idea for the long run at all

- there doesn't seem to be much headroom in the GT72 for OC unfortunately because of the cooling solution and the ec-controler

- for the GT72 we might need a "optimized throtteling" bios that does not cross healthy limits with a much lower power target

OK... Did another few runs and here is what I found

With prema bios at stock clocks:

- the total power readings of the "MSI dragon gaming center" seem to be quiet accurate (I compared them to readings of my power meter)

- the power consumption with premas bios is exceeding 245 Watt

- with the stock ac adapter the GT72 is constantly switching to "hybrid" power - meaning that the battery is used in addition to the AC adapter to compensate the high power demand (this can also be seen in "dragon gaming center")

With svl7s bios at stock clocks:

- same story - in addition I even needed to throttle the power limit slider down to 50-65% - otherwise I got to a point where the GT72 throttled itself (the internal EC controler)

With OC (both bios mods)

- before I redid the paste job I also saw a lot of throttling when using any OC due to overheat (exceeding 87 °C)

- with svl7s bios I needed to up the voltage slider higher than with premas mod to be able to OC (also I assume that this is due to different base lines - I didn't compare the actual resulting voltages)

- overall OC seems to be possible with both but there is much more trial and error with svl7s bios and GT72 owners seem to need to lower the PT slider A LOT - lowering the temp target does help too but gives much worse overall scores / results

- the power consumption in dragon center starts to show "AC" instead of any actual value during test runs - this might indicate that it is exceeding all limits

Bottom line for GT72:

- a stronger AC adapter might be needed - the GT72 seems to have "dell compatible" connector - I sucessfully tested a 240Watt ac adapter from dell - but it's actually shutting down a little earlier than 240 Watt (might be because its old) to protect itself compared to the MSI one , question is if the GT72 will detect the stronger power source and act accordingly

- with the GT72 switching constantly to hybrid mode and thus also decharging the battery and possibly using up more recharge cycles I'm wondering if OC is a good idea for the long run at all

- there doesn't seem to be much headroom in the GT72 for OC unfortunately because of the cooling solution and the ec-controler

- I will still need to check if the stock bios is requiring hyprid power as often as the modded bios (UPDATE: See update)

- for the GT72 we might need a "optimized throtteling" bios that does not cross healthy limits with a much lower power target

UPDATE1:

Stock Bios - Stock Clocks:

Did a few test runs with the stock bios and it seems not to exceed 210 Watt whereas the OC bios reached 245Watt (and probable more depending on what "AC" is meaning in dragon center).

In respect of hybrid power I saw it only triggering once for a short time during my 3D mark runs.

UPDATE2:

Stock Bios - Max OC (which isn't that much +135Mhz GPU, +600Mhz Mem, +12.5mV).

Hybrid power triggered more often and also for longer periods, yet still less often as with the OC mods (even if those run at stock clocks). 220 was the highest Wattage I've seen.

Question is - why does Hybrid Power trigger at all - isn't this supposed to be a 230 Watt ac adapter (19.5Vx11.8A) ?

..so there's no way to flash for me , alienware m17x r4 120hz, infinite sadness...however I have windows 8.1 and hidevolution 980m whit their own drivers, 8400 at "fire strike", no overclock. Is it good? How to know if my card is throttling?

Sorry..quite noob on the argument.

8400 to 8500 is what you can get without OC - so that's "fine" for a stock bios...

But even with a stock Bios you should be able to OC by +135Mhz Core and 600Mhz Memory and that should give you a little boost to somewhere in the 9k range...

That UEFI thing is really strange. Up to now I was convinced that this is "just" important for booting etc. - didn't expect it to influence nvflash in windows. Even more so since my GT72 is in UEFI mode too and I can still flash. I just can't understand how the ability to theoretically switch to legacy mode is influencing what is possible in UEFI mode... Is there any reason to believe that Alienware didn't just lock out the option to flash intentionally and this has more to do with Alienware and less with UEFI ? But I'm here to learn and that's what I do - next time I will remember that there are issues with certain Alienware laptops and flashing the vbios...

You are not reading what I wrote. Read:

You obviously have Legacy enabled because you were able to use mnvflash. It doesnt work with Legacy disabled and UEFI enabled. Thats pure UEFI.

You understand the difference?

Alienware 18 with 970M/980M can only use pure UEFI. No legacy, no mnvflash or nvflash.

GT72 can use whatever it wants. Legacy or UEFI or both

No... I do not... Because from all I know it does not make a difference... he needs to use the modded nvflash (mnflash) and that is a windows programm... it shouldn't and doesn't care about uefi... for all I know...

He can't flash because he still is not daring to deactivate both cards at once... I'm quiet sure that's why...

For the rest... For all I know there is no "both" - it's either uefi (with or without secure boot) or legacy... some can switch but no way to use both (at the same time)

It won`t work.

You have UEFI and Legacy enabled.

We have UEFI and Legacy disabled.

mnvflash only works with Legacy. If we enable Legacy we are fucked so we can`t use that option.

Disabling 1 or two cards, deleting drivers, running in windows, using cmd etc etc, it doesnt matter. UEFI are in the way both mnvflash and nvflash doesnt support pure UEFI.

I'm sorry but how do you know what "we others" have ? My GT72 is in UEFI mode, windows was installed in UEFI mode... In my opinion this does only matter if one wants to boot and flash in dos mode...

Furthermore if he had actually legacy enabled like you are saying - that wouldn't be PURE uefi - that would be legacy (=bios) mode... And then I would wonder how uefi could be in the way...

Which is not necessary... However I'm out of that discussion now as there is not more info I could provide...

I think this is the reason why NVidia disabled the oc features in mobile drivers especially for 9xxm cards

Check whether your card can use all of its VRAM

@Rhadamantis..

I'll check where I put it.

Honestly - the gtx980m seems to be quiet a beast. I have a second ac adapter from dell I use for travelling because its smaller - its rated 240 watt - it has repeatedly shut istelf off during benchmark runs with +250 Mhz Core and +50mV today. This is telling me that this ac adapter is either defective or we should actually be careful. The GPU might be able to take this much power and it is kept cool by the heatsink - but remember that the power needs to run through traces that have been designed for less power. There certainly are tollerances but I'm left wondering how much more power an unlocked, overvolted and overclocked gtx980m is actually sucking compared to a stock one... I will need to find my power meter...

Of course.

That's bad news... So they days of modding and occing on laptops might be coming to an end.

Nvidia inspector is able to detect a modded bios and so would be nvidia. So for some time we might still be able use a modded bios but I assume that days will also end in the nearer future .

According to all I've read in the last weeks: You will need to disable BOTH cards or otherwise the nvidia driver won't deactivate and the modded nvflash won't be able to access (or even find) even a single one card...

If you do not dare to disable both cards simultaneaously also multiple people told you that nothing bad will happen you could try by uninstalling the nvdriver, reboot and run the nvflash with NO nvidia driver running in windows.

Because that is the goal of all this - the nvidia driver MUST NOT be running or nvflash won't work at all (at least that's what I got from all of this...)

All 347-series driver prevent OCing via software on mobile Maxwell...

Uh oh... are they preventing it "on purpose" ?

But its useless. I tried disabling the drivers, SLI on and off, with drivers, the mnvflash still cant find any cards. Its because it doesnt support UEFI. So we are stuck

I posted screenshot a few pages back

It doesn't need to support uefi because you need to flash in windows and not in dos... My machine is using uefi too and I've no issues. If you do not want to disable both cards at once you can try to uninstall the nvidia driver and then try to flash. Other common mistakes seem to be: You need to run the command shell as admin. You need to use the nvflash version provided here and no other version.

PS:Please check your - post it's looking wrong ...

Something is shuffling the quotes...

New findings on the GTX980M in my GT72 with this mod... Still following my idea that something is throtteling with Svl7s bios and the MSI GT72 and after reading his OP again I flashed it again and did a few runs at the default clocks of his mod and just played with the power target...

I'm able to OC memory to +600. I did that to be better able to compare my results with stock oc and Prema oc as I used that mem OC there too for most of my runs...

I did runs with putting the PT slider more and more down and found my 3D mark firestrike score improving the more I turned the PT down...

If someone wants to verify my findings please disable "priotize temp target"...

The sweet spot for my GT72 seems to be at about 68% power target... That's where I'm able to archive about 9200 total score in firestrike at the "stock clocks" of the Svl7 bios.

Update:

By using a voltage increase of 87.5mV I was finally able to also OC my GPU with this vBios by +100Mhz. Result was a 3d mark firestrike total score of 9500 which is getting closer to what is possible with Premas vBios. I assume without knowing the baselines for both mods (power target at 100%, voltages, etc.) it's hard to come to a final conclusion... So OCing IS possible and I think with enough fine tuning it's possible to get comparable results with both vBios version. I'm just not sure if they would endup beeing archived in the same way or if there is something that can be learned here...

I'm just hoping that the modders here can maybe get some ideas of my/our findings and give us a "best of both worlds" modd

a modded ec firmware for gt72 ? where did you find it ? and we have the same model with same specs..

anyway there is bios update for gt72 ==> MSI Global - Computer, Laptop, Notebook, Desktop, Motherboard, Graphics and more

says that : Modify ME VSCC table support BIOS ROM:GD25B64B we need to update with this thing or ?

btw im planning to activate super raid 3 but do i need to buy one more 128gb ssd to do that or 2 ? i have 2 pieces of 128GB Kingston SSD on my rig right now

and did you overclock the cpu i7 4720hq? i did with xtu +200mhz but cant see a worthy improve

- - - Updated - - -

so thats true we have to cooperate for making a custom vbios together, BigKid...! i dont know using hex editor btw

I will answer your questions about EC firmware modding and Raid in a PM - do not wan't to go Off Topic...

@BigKid I dont know what will happen if i deactivate both cards, cause i just have my 2 980ms in Graphic cards list.

Well I assume the same as with others that have only one... I can deactivate the card in my GT72 even when I'm in "dedicated gpu mode" (the GT72 has no optimus but you can switch between intel hd and nvidia (requiring a reboot)).

As far as I understood it the reason why one needs to deactivate the card(s) is to get the nvidia driver to deactivate and windows to switch to a default fallback driver so nvflash is able to access the cards... And I assume that this won't happen if you do not disable both cards...

It's up to you. As MSI GT72 owner I've tried to follow all this very closely and according to my understanding the "genral" issues with flashing a clevo vbios to a msi card are gone since the last generation MSI of notebooks (starting with GT70). So there might never be a "msi" vBios because there is no need/reason for it... So the best I do currently hope for is that we get a clevo based SLV7 one that is also working on our GT72s.

I do not want to go OT. Besides comparing SVL7s to Premas Mod to find out what might be causing the issues GT72 owners have with the SVL7 mod but NOT with the Prema mod I'd like to follow Premas wish and forum rules and focus on SVL7s mod here and all I add is - I do not know how you came to the conclusion that there are issues with GT72 and Premas mod because there are no known issues (besides the missing power target slider and that's just ME - I'd prefer to have one) - please ask there in case you want to discuss that...

From what I last read I`ll rather wait for an dedicated vBios for th MSI GT72

I`d be happy if there`d be one some day.

Best regards and an extra big "Thank You!" for the efforts of the modders!

phila

Well you might increase the chances if you give the existing ones a try and experiment a little and share your experiences.

This is what I found out so far on my GT72 system:

Premas Mod is working and gives good OC capabilities. However my machine (not my ac adapter) is making a mild humming sound while running 3D apps. I do not want to get OT - I'm just mentioning it as a reference and pointing out the differences. Prema doesn't offer a power target slider and maybe that's part of why it is working better on our machines. It also does not allow undervolting yet.

SLV7s mod is working at stock clocks and with the power target set to 50%. It gave me 3dmark scores comparable to stock then. There are reports that it throttles on GT72 with PT set to 100% even at stock clocks (most probably it's the EC throtteling because of total power draw, not the GTX980M because of its power draw) - I haven't tried that myself.

Even the slightest OC is not woking on my machine with SLV7s vbios (I tried +100Mhz, +50mV, and several power targets (50,60,65,70%) without any luck.

So two (suspected) issues:

1) Probably: The power draw of the gtx980m get so high the whole system throttles down (so if slv7 could provide a version with a lower default PT we could try that)

2) There seems to be a wall we are hitting when trying to OCC - currently no idea - is the default voltage decreased so we are using a too low offset and would need to go beyond +50mV ?

I am not able to list a card. I tried to deactivate 1 of them, but it is not working. mnvflash --list -> Nothing happens... So i cant flash them...

You might have to deativate both before using this nvflash version...

i looked at your 3dmark score and saw smthing interesting your overall score is very close to me mine is 10197 overall but your graphics score 12373 but mine 13083..?

your other score is better than me so you get 10169 even we have the same cpu i7 4720hq

my values are 1365mhz gpu 6000mhz mem 1.1volt(+100mV) causes to make max 76C, even in watch dogs i reached 83C then with cooler boost reduced back to 72C no problem..

and your +50mV is not enough to feed to gpu in real time test just like playing games and you get crash or freeze with your values, dont trust 3dmark

You might be right but my experience is that with -10, -20 Mhz adjustments I'm typically able to stabilize it from 3Dmark stable to game stable.

However in respect of the variations of our 3D mark scores: I'm using a modded ec firmware. I modded it in a way that the fans are kicking in later and less violently to get an overall more silent system.

The drawback is that this way my card stabilizes at about 80-85 degrees wheras the stock ec firmware is stabilizing the GPU in the 70-75 degree range... More heat -> more power needed -> slightly lower result... That might be one reason. I also do not know why my overall score is better when the grphical score is worse - I do not know if the sdds are making a difference (using a 2xSDDs in RAID0) or mabye because I'm using less power for the GPU (only +50mV vs your + 100mV) I've more power left for the CPU before the throttling by the EC controller starts ?

With my AW18 980m SLI (only UEFI) its not working to flash with mnvflash... I think its locked cause of the BIOS. I hope svl7 will find a fix.... PLEASE ^^

Did you to to flash using i 0 and i 1 ? If you have SLI you need to tell the nvflash which card you'd like to flash...

After testing Premas vBios also wanted to test SLV7s vBios also I knew that there are issues with the GT72 and this vBios. However I would still prefer to have influence on the power target so I'd prefer to use the slv7 version.

(yes - I'm a little stubborn here but I consider it a safety net even when I stand corrected that it might not be a accurate one - I do not want it to do more occ but actually to limit GPU boost a little to not overdo it).

My System is a MSI GT72, GTX980M 8GB and I'm using nvidia driver 344.75 and nvidia inspector to OC.

This are my findings. To be able to complete a 3dmark firestrike run I needed to set the power target to 50% (!) at stock clocks. That gave me a 3D Mark score comparable to an 100% stock run (about 8500).

As soon as I raise the clocks I can not even complete a few seconds of 3D Mark even with higher voltages (tried +50mV for +100Mhz Clock, I tried multiple PTs 50, 60, 75%)

With stock bios I can run +135Mhz with a +25mV increase.

I do not know why I'm hitting a wall with any OC with this vbios (I can do a lot more with Premas bios so it's not the chip or the GT72 in general).

However I've a guess why GT72 owners are getting lower scores or stuttering with the GT72 and this vbios. Maybe the power limit has been set too high - so the actual power draw of the GTX980M gets so high that the overall power draw (CPU + GPU + everything) in the combined tests exceeds the laptops limits and thus the GT72's EC controler throttles the whole system. I'm thinking this as I could get comparable scores to the stock bios with setting a 50% power target with this OC vbios. I know that someone watched the MSI Systems monitoring tool and it was indicating 175Watt Power draw when the system throtteled but I'm wondering how accurate that tool is and if it is considering all other parts that need power too (like keyboard lighting, SSDs, HDD, displays backlight etc etc etc...) which the EC controler might actually be considering / measuring...