Mr. Fox's GeForce 345.20 Desktop Driver Mod for Alienware and Clevo Mobile GPUs

Mr. Fox's GeForce 345.20 Desktop Driver Mod for Alienware and Clevo Mobile GPUs

OK, I know everyone has been eagerly waiting for another one of these. This 345.20 driver appears to provide near identical performance to my last 344.75 desktop driver mod, so here is it. I hope it helps 880M stability and performance as much as the last one.

Tested on Alienware M18xR2 with 780M SLI and Clevo P570WM with 980M SLI. I have added broader GPU support this time. This uses the desktop INF and requires the hardware ID be removed from the mobile driver INF. A modded desktop INF is included (brand agnostic), modded Dell/Alienware INF and modded Clevo INF are also included in the package.

If you want to use this mod on an ASUS, MSI, Lenovo or something else, it should work fine, but you will need to find your brand's INF and remove your hardware ID from that file to force it to use the NV_DISP.INF as I have already set up for Alienware and Clevo.

Supported GPUs:

• NVIDIA_DEV.1198 = "NVIDIA GeForce GTX 880M"
• NVIDIA_DEV.1199 = "NVIDIA GeForce GTX 870M"
• NVIDIA_DEV.119A = "NVIDIA GeForce GTX 860M"
• NVIDIA_DEV.119F = "NVIDIA GeForce GTX 780M"
• NVIDIA_DEV.11A0 = "NVIDIA GeForce GTX 680M"
• NVIDIA_DEV.11E0 = "NVIDIA GeForce GTX 770M"
• NVIDIA_DEV.11E1 = "NVIDIA GeForce GTX 765M"
• NVIDIA_DEV.11E3 = "NVIDIA GeForce GTX 760M"
• NVIDIA_DEV.1211 = "NVIDIA GeForce GTX 580M"
• NVIDIA_DEV.1212 = "NVIDIA GeForce GTX 675M"
• NVIDIA_DEV.13D7 = "NVIDIA GeForce GTX 980M"
• NVIDIA_DEV.13D8 = "NVIDIA GeForce GTX 970M"

As usual, On-the-Fly-SLI is ready to go and I have included @j95's CUDA Encoder mod... thank's Brother J.

Notice graphics scores are near identical to 344.75 and 344.80 in these examples... well within the margin of error.

Alienware M18xR2 780M SLI - Overclocked - 345.20

Alienware M18xR2 780M SLI - Overclocked - 344.75

Clevo P570WM 980M SLI - Stock - 345.20

Clevo P570WM 980M SLI - Stock - 344.80

GTX 980M Overclocking Support (example)

j95's CUDA Encoder Mod Included


So... get started by downloading the DESKTOP driver: [ 345.20 WHQL Desktop GeForce Driver]

• Extract the desktop driver package to whatever location you want it
• Download this zip file [345.20 Desktop Mod Files]
• Extract the contents of the zip to the Display.Driver folder, overwrite existing files
• Run DDU in Safe Mode to get rid of any old drivers and registry entries
• Run Setup.exe, answer "yes" to unsigned drivers, reboot when done
• Enable SLI (if applicable) and set up NVIDIA Control Panel to your specs

Reminder: Windows 8.X.X users, remember to disable driver signature enforcement prior to installation or it is guaranteed to fail.

This was only tested on the machines mentioned above. You assume the risk of any complications that may arise if you choose to use the driver mod on any machine.

For your convenience, I am including this video here so you can set up ShadowPlay desktop capture if you want it. (Doesn't work with Optimus.)



In case anyone has an issue accessing the mod file, joluke has an alternative link (see below). Thank, joluke.

---

***ATTENTION WINDOWS 10 USERS WITH 780M SLI THROTTLING ISSUES***

---

Try this. It works and I just recently discovered the command.

Open an elevated command prompt and run the syntax below. Reboot, and DSE should be disabled for good. No need to temporarily disable DSE the old way.

Code:

bcdedit.exe /set nointegritychecks on

I am now able to install modded drivers without having to boot with DSE temporarily disabled. Works on the Sky X9 and Panther with W10 TH2. I've done it probably a dozen time in the past two weeks.

 

These work well enough for me in the R2 to keep. Ill just add DDU was necessary to get this to work. I didn't use DDU the first time and I got a some strange image doubling.

Cheers for the mod including Brother J95's CUDA hack too and the Shadowplay tutorial

 

Are you by chance using a 980M ? This symptom you mention of doubling, seems to be consistant with another user i ma helping, who has issues with an 18 r2 and video issues when changing drivers. Though he was not using the previous desktop mod driver, but rather 344.91, and couldn't roll back to 344.75 or even complete an install of it. Now he is stuck with some black screen, when I remote into his PC I can actually make some images when i move the mouse around, as is the mouse was an eraser. Any idea's what to do next ? We've exhausted a few things..

Anyways - to those who try to use this type of driver with 880M gtx's, I remind you, do not rob all your graphics power by overclcoking everything else, or having 20 things plugged into every USB port and every HD port had port multiplied and are using mechanical drives..

About 5 of you guys who messaged me with questions and how to's all had the same issue, being that you were robbed of power. Power lacking causes throttling, and increasing the volatge is NOT always going to help you NOT throttle, in fact, if your out of power, increasing the voltage might make you crash, experience non-stop unpredictable throttling, or even make other components look as if they are failing.

Also - you don't always need a physical number to to be highe ri.e 4ghz vs 4.5ghz... to get more performance.. to get stable at those higher frequencies, requires exponentially more voltage. you also are less and less in sync with your FSB multipliers as well, menaing you lose more performance then gain potentially.

I can match any of your CPU benches and scores with usually 500mghz or less overclocked and even less voltage sometimes. I/O's only need a logic high or logic low to know if a bit or byte is being moved/flipped/inverted/changed etc.. logic high and logic low on NAND IC is very low, on 22nm cpu's I can only image how low it is. Increasing the voltage does not make things go faster, but rather eliminates waiting time for a capacitor to charge. If it charges too fast, it can think it did not discharge, and an error can be created, and it will produce incrediable heat as it tries again, to dissipate. Things also latch in rows and columns meaning its more then just one bit or byte at a time..

To find your stable clocks, I suggest you clear a schedule for the day, cause it takes time. ALSO - people these days forget that there are BUS Holes inbetween ratio's of surface mounted and inbedded circuits and processors... like the RAM on a videocard needs to talk to the GPU itself... there sometimes exists these holes where increasing the voltage yeilds little gain in stability or even throughput. A usual clear sign is when increasing the voltage to something (keep in mind very veyr small incriments, like the smallest possible in toggles) , gives no more performance gain (keep in mind, im saying that without even increasing the speed/mghz).. thats when you hit the bottom limit of a hole - now you need to find when it starts to make a difference again. Then make note - any voltage in that region causes heat and no gain in performance. Then try getting the voltage to the higher limits before a hole, and increaidsng the Mgz if you want real performance.

I find the last 344.75 driver was good for this. I was able to get ore performance by lower my clock speed and keeping stable/stock voltage, essentially I eliminated heat, and wasted cycles of refresh. Also, remember voltage comes with its own speed, in Hz...think of it as a pool, and every cycle of processing takes abit of the pools substance away, until it refreshes at set intervals with its own amount. Knowing this, you can understand why you need to increase the voltage when you increase the speed, because it becomes an unsustainable equation. BUT** you can always expect some performance gain, by just increasing the voltage, instead of the speed too, and usually this is the refresh that happens faster.

Have you ever seen a fast CPU with very little cache ? sometimes it feels like it stutters for a split second but then everything can happen all at once...even as CPU's get faster but now throttle more, their amount of time to react is becoming larger, this is like the "ping" on your internet connections even, if you understand what im getting at. Its the factor that determines how nice things look on screen in many ways, thus its like saying its graphically preferable if things react/ping/refresh faster.. increasing the voltage on things without increasing the speed (mhz) usually results in this;. You can do this to your RAM, to your CPU, to your GPU..etc.. and technically not be over clocked, but getting the same performance as being over clocked.
If you then increase your speed afterwards - when your at the max voltage that makes a difference, just be careful not to increase the speed so much that your depleting faster then restoring/refreshing. In fact, increasing the speed, can change the Hole's position, but its hard to figure out by how much, though it surely is all relative. Experimentation is key.

I have also never seen , in 18 years of overclocking CPUs and otherthings as they came around, burnt something out, or destroyed it from overclocking. I've destroyed things by not cooling effectively, by spilling liquids, short-circuiting, physical damage, but never from too high voltage and speed. Now, i am not saying go nuts, I am saying, do not be too shy to find the NEXT lower and upper limit past the HOLE in performance that you see. These holes are, for the more understanding of people, limits in ratio's that make no sense, and differences in I/O's being considered logic High when they should be low. By the latter - I mean to say, sometimes, an INSTABLE, voltage can be made STABLE by actually increasing the speed too, kind of like preventing the "POOL" from overspilling too. So, its all a balancing act.

As work needs to be done, you need to make sure the force needed is present, and in enough supply, and ever replenishing constantly, and that when there is nothing to do, nothing overflows...meh hope my rambling helps.

i also like the fact Fox's driver was stickied and indexed well enough I can type 344 fox and found a link tot download it haha !



Here are my results using this driver right now, with an m17x-r4 laptop.

http://www.3dmark.com/3dm/6035323?

apparantly, this computer:

• Intel Core i7-4770K
• NVIDIA GeForce GTX Titan
• ASUSTeK COMPUTER INC. Z87-PRO

is only 3-4% higher then my:

• Intel Core i7-3940XM
• NVIDIA GeForce GTX 980M
• Alienware M17xR4
• 64-bit Windows 7 (6.1.7601)

So technically speaking, its cheaper to have higher performance in a laptop then in a desktop.
This laptop, even with the two samsung 850 pro ssd's and the 980M gtx, and 1080p screen only cost me like 2,000$
Keep in mind I get a gaming keyboard and decent monitor and PSU and speakers and mouse and all that, included...


The number 1 place with same CPU and GPU is 300 points below me. My score will not rank because the drivers are unrecognised, but just to give you the example, that this driver OBVIOUSLY works great, and this time I did NOT use a 880m GTX, I did a single quick test with my m17x-r4 and 980m


**********

LOL Fox, you have such a deep deep voice haha

@

 

adding this to the batcave OP thx fox!

 

Thanks for the mod, I've intalled it right now. I will test it asap.

 

Does anyone know how to achieve CUDA video encoding with Sony Vegas Pro 13? Thanks

 

@5150Joker might know that. I think he uses Vegas a fair amount. As the imaged showed, I use Nero.

 

@Mr. Fox great drivers thanks

http://www.3dmark.com/3dm11/9484360

 

@Mr. Fox dunno if you already know this, but 345.20 is simply 344.75 with patched nvlddmkm.sys and nv4_mini.sys to prevent some kinds of driver modding after Nvidia got scared from the whole mobile G-Sync business. That's why you noticed no performance difference from 344.75.

 

Nope, didn't know it. Ah well... as long as it works good, who cares, right? 344.75 is a decent driver, too.

 

I haven't used Vegas in a while but basically when you go to "Render As" during the final steps, open up a custom template and under the video tab at the bottom is a drop down for encode. If you have CUDA enabled (usually installed by nvidia drivers), then you can simply choose "render using GPU only" and it should take care of it.

 

345.20 includes v344.80 hotfix too, but with v344.75 performance instead.

 

I̶'̶v̶e̶ ̶n̶o̶t̶i̶c̶e̶d̶ ̶a̶ ̶p̶r̶o̶b̶l̶e̶m̶ ̶a̶n̶d̶ ̶I̶'̶v̶e̶ ̶h̶a̶d̶ ̶i̶t̶ ̶r̶e̶p̶l̶i̶c̶a̶t̶e̶d̶ ̶2̶ ̶t̶i̶m̶e̶s̶ ̶s̶o̶ ̶f̶a̶r̶.̶ ̶F̶o̶r̶ ̶s̶o̶m̶e̶ ̶r̶e̶a̶s̶o̶n̶,̶ ̶m̶y̶ ̶m̶a̶c̶h̶i̶n̶e̶ ̶h̶a̶s̶ ̶b̶o̶o̶t̶e̶d̶ ̶w̶i̶t̶h̶ ̶a̶ ̶B̶l̶a̶c̶k̶ ̶s̶c̶r̶e̶e̶n̶ ̶2̶ ̶d̶a̶y̶s̶ ̶i̶n̶ ̶a̶ ̶r̶o̶w̶.̶ ̶ ̶E̶a̶c̶h̶ ̶t̶i̶m̶e̶ ̶h̶a̶d̶ ̶a̶ ̶n̶e̶w̶ ̶c̶l̶e̶a̶n̶ ̶i̶n̶s̶t̶a̶l̶l̶ ̶w̶i̶t̶h̶ ̶D̶D̶U̶ ̶w̶a̶s̶ ̶d̶o̶n̶e̶ ̶t̶h̶e̶ ̶n̶i̶g̶h̶t̶ ̶b̶e̶f̶o̶r̶e̶.̶
I̶t̶'̶s̶ ̶v̶e̶r̶y̶ ̶s̶t̶r̶a̶n̶g̶e̶ ̶s̶i̶n̶c̶e̶ ̶I̶'̶v̶e̶ ̶r̶e̶s̶t̶a̶r̶t̶e̶d̶ ̶s̶i̶n̶c̶e̶ ̶i̶n̶s̶t̶a̶l̶l̶i̶n̶g̶ ̶t̶h̶e̶ ̶d̶r̶i̶v̶e̶r̶s̶ ̶e̶a̶c̶h̶ ̶t̶i̶m̶e̶,̶ ̶a̶n̶d̶ ̶b̶e̶n̶c̶h̶e̶d̶/̶g̶a̶m̶e̶d̶ ̶a̶ ̶b̶i̶t̶ ̶t̶o̶o̶,̶ ̶b̶u̶t̶ ̶a̶f̶t̶e̶r̶ ̶a̶ ̶s̶l̶e̶e̶p̶ ̶(̶y̶e̶s̶ ̶I̶ ̶h̶a̶t̶e̶ ̶n̶o̶t̶ ̶m̶a̶k̶i̶n̶g̶ ̶s̶e̶n̶s̶e̶ ̶t̶o̶o̶)̶ ̶o̶n̶ ̶a̶ ̶c̶o̶l̶d̶ ̶b̶o̶o̶t̶ ̶I̶ ̶g̶e̶t̶ ̶a̶ ̶b̶l̶a̶c̶k̶ ̶s̶c̶r̶e̶e̶n̶.̶

I̶'̶m̶ ̶t̶r̶y̶i̶n̶g̶ ̶t̶o̶ ̶m̶o̶b̶i̶l̶i̶t̶y̶ ̶d̶r̶i̶v̶e̶r̶s̶ ̶o̶u̶t̶ ̶n̶o̶w̶.̶.̶ ̶w̶i̶l̶l̶ ̶s̶e̶e̶ ̶h̶o̶w̶ ̶t̶h̶a̶t̶ ̶g̶o̶e̶s̶.̶

UPDATED BELOW

 

we've been trixed ! o.0

 

I tried this out last night and it works perfectly fine with my finicky 880M. Thank you Mr. Fox!

 

Did you get it fully sorted? What is BIOS settings?

 

I also install and works perfectly on my AW17 with GTX880M.
The performance is very good in my GPU.

 

When using latest DDU v14.0 uncheck AMD Audio Bus (bug), setting doesn't stick so you will need to repeat this each time. Restore defaults PEG0,1.,2,3 & SA DMI configuration BIOS tweaks, those are for the 120Hz AWs (PEG0).

@Mr. Fox included profiles?

 

So that AMD setting is the cause of it?

 

Heres a new desktop driver for you to mod Fox!!

http://www.guru3d.com/files-details/geforce-347-71-driver-download.html

 

Is it another clock-blocked turd driver or an unlocked one?

 

Clockblocked. It's the same as 347.52 with one additional hotfix: http://nvidia.custhelp.com/app/answers/detail/a_id/3636

 

Thanks... I thought so. In that case, I won't be testing or modding it. I don't waste my time tinkering with filth.

 

Yeah, let's wait for the proper 349.xx driver... It better come soon

 

Woops I should have known it was too early and too good to be true. We'll have to wait it out a little bit longer.

Just tested out this driver - Thanks Fox! Your right its performance is roughly the same as previous drivers.
So far for me the 347.26 (Clockblocked) stock GPU clocks give me the best results in terms of stability and performance. So I hope that the new unblocked Nvidia drivers next month start from there and then having the ability to OC again we will be able to boost scores even higher.

 

Just wanted to update this. It appears my issue was related to a crapped out BIOS (just a coincidence). Who would have guessed huh..actually Mr Fox kinda of did. Cheers Fella. Long story, Short, I fixed the problems and am now using these drivers without a hitch.

Ohh and have that CUDA support shaved loads of time off a video render I did in Nero. Just what the doctor ordered as I need to use Nero over the next couple of months

UPDATE #3

Nope, something ain't right with these and my system. It's the dame for the mobility drivers too. I get a black screen after a reboot sometimes and have to go into safe mode etc... then uninstall.

@j95 and @Mr. Fox

I want me some CUDA support for Nero. Is it difficult to add it in ? I'm back on 344.91, perhaps could learn how to become self sufficient.

 

Extract w/7-Zip both nvcuvenc64.dl & nvcuvenc32.dl -> rename "nvcuvenc64.dll & nvcuvenc32.dll" to "nvcuvenc.dll".


nvcuvenc.dll (nvcuvenc64.dll/) -> copy to Default/System32.

cmd (Admin)

Code:

 %windir%\System32\regsvr32.exe %windir%\System32\nvcucenc.dll

nvcuvenc.dll (nvcuvenc32.dll/) -> copy to SysWOW64

cmd (Admin)

Code:

%windir%\SysWoW64\regsvr32.exe %windir%\SysWoW64\nvcuvenc.dll

-> reboot.

#3 http://forum.techinferno.com/nvidia/8290-[mod]-latest-nvidia-drivers-cuda-nvcuvenc-encoder-support.html#post114350

 

Cheers brother this is awesome. Thanks for taking the time.


There's a few things I'm not sure if I follow.
What am I do do with the all the CMD promt addresses...?

Code:

 %windir%\System32\regsvr32.exe %windir%\System32\nvcucenc.dll

and

Code:

%windir%\SysWoW64\regsvr32.exe %windir%\SysWoW64\nvcuvenc.dll

What am I do do here? I am supposted to run the dll's? If so I am unable to run anything from CMD after I've dropped the files in SysWow64 and System32..?

Also step 3 ?

Code:

[nv_commonBase_addreg]
HKCR,"Wow6432Node\CLSID\{B63E31D0-87B5-477F-B224-4A35B6BECED6}",,%REG_SZ%,"NVIDIA CUDA Video Encoder"
HKCR,"Wow6432Node\CLSID\{B63E31D0-87B5-477F-B224-4A35B6BECED6}\InProcServer32",,%REG_SZ%,"%10%\SysWow64\nvcuvenc.dll" 

?

 

No, if I add the file that contains the newer profiles from clock-blocked crap drivers to this driver package it fails to install in Windows 7 and 8.1.1 so they will have to be manually installed or installed through GFE after installing the modded driver. Probably a version incompatibility issue.

 

Just copy the renamed nvcuvenc DLLs and then run the corresponding commands to register each nvcuvenc.dll. There's no step 3, it's just a reference/INF.

To clarify some things:

1- v347.xx are the best drivers for 900s so far, probably one of the many reasons NV blocked OCing in the first place 'entiende', nvidia must remove the clock-block without messing up v347.xx TDP or it will be a throttling nightmare for AW/PEG users.

2- HotFix drivers include many fixes, most of which NV won't admit + internal testing. e.g. v344.80 unrelated 'official' fixes and inferior performance compared to v344.75. NEVER THE SAME!...

Yep, or the patched nvlddmkm.sys interfering.

 

@Mr. Fox here you have a mirror for your modded drivers with the already modded files inside Display.Driver:

It's exactly like your other mod, extract and install:

(don't forget to disable signature verification of course)

https://mega.co.nz/#!88p2iQ5A!IL_KBn7L80IMFIEqrYKrTizepvR6M7bi8GzBBOSz8xs

Btw, feel free to add it to OP

 

Hmm thanks @Mr. Fox but I'll probably wait until 347 comes out with your modded files.

 

No 347 mod coming unless they release a 347 without any Maxwell clock-block garbage.

@joluke - thanks for the mirror

 

AFAIK they will release next version without clock-block

 

You're welcome my friend

 

They said so.. I believe it when I see it .. happy with my frankenstein modded 346.87/347.52 drivers currently

 

*svl7 & Johnksss* vBIOS MOD - Alienware/PEG GPU throttling...

 

... What's going on here brother. .? Was this the method you outlined at TI?

 

svl7 & Johnksss vBIOS MOD + my driver mod + ....PM. Gone...

Spoiler

 

nice, that is indeed great news for our alienware brothers! good job there j95

 

Hours later, guess what...

Spoiler

 

Do share.... I didn't understand your previous message about the "PM.. Gone"... I was looking in my inbox..?
But if you're found a non throttling solution I'm all ears
Please explain.

 

Yeah don't see the point of not sharing..

 

Maybe @j95 wants to confirm it is a real fix before sharing, and maybe he wants to be careful about posting things that Alienware or NVIDIA can use to re-screw us down the road. In other words, it's harder for them to interfere if they don't know what is being done to sidestep their crippling shenanigans. We already know that ManuelG is trolling our forums and Tech|Inferno. We don't know what he is doing with the intelligence that is gathered. We can only assume it's not honorable.

I am all for sharing, but it might make good sense to taking the sharing underground on something like this (where tweaks are circumventing driver stupidity) and only sharing by PM with people you know are not spies from NVIDIA or Alienware. At least for a while keep it on the down low among an exclusive group of well-established and known trustworthy forum members that have machines which suffer from cripple tactics so that NVIDIA and Alienware have no information to work with other than seeing the code has been broken and crippled machines are finally working as they should. Those that have machines that function correctly with 980M (i.e. some Clevo models) don't need to know the fix. They have nothing that needs fixing.

 

aside from the clockblocked vbios, but prema already took care of that

 

Exactly, no hints whatsoever...Gaming & GPU intensive applications throttling is gone, unlike clock blocked v347.52 occasionally dropping 1113 MHz. Serious OCing I was able to achieve 12880 graphics, better than many scores out there without any throttling/ core & memory fluctuations, IINM even @Mr. Fox's best M18xr2 results core was still fluctuating? This works but it needs to be applied differently for the so called bad days when OCing (3), even those it won't throttle like before, meaning gaming without any FPS drops/ stuttering. Clearly this was meant/intended not work with previous AWs, which is extremely complicated to achieve without messing with everyone else.

 

Mr. Fox, can do the mod. driver 349.65 Desktop for Alienware?

 

I don't know. That's the Windows 10 WDDM 2.0 driver. I don't have Windows 10 on any systems any more, and I am not convinced it's worth the "free" upgrade. Micro$haft is still way off the mark and it's not as good as Windows 7. I tested Windows 10 for a while, got really bored and was quite disgusted with what a pile of crap OS it is. I'm not sure the driver is structured the same as far as INF is concerned because I haven't recognized a need to look at it yet. Maybe @j95 has already taken a peek at it. If not, maybe I will look at it this weekend for you.

 

349.65 sounds like a total lemon from comments I've read.

 

I use Windows 10, for me it is a very good OS. I will be very grateful, if you make a mod for 349.65. Thank you Mr. Fox.