NVIDIA Kepler VBIOS mods - Overclocking Editions, modified clocks, voltage tweaks

[Nofew]

Decoy101x: That's called your "boost clock". Basically, let's say you got an nVidia 680m. The stock clock on it is 720 mhz. No matter how high you set Afterburner, as long as it's over that, the first thing it'll try doing is running at 720 mhz. If it's fully stressed out (thus clocking faster will benefit framerates), it'll then "boost" up to what you /actually/ have it set to. It's possible the "base clock" will be different; it's possible to set it in the vbios, but the general concept is the same.

The constant jumping is happening because you're running the stock bios. I don't know why for sure, but a vbios does fix it while still retaining the "boost" feature. In my case it was even worse; It'd go from 888 mhz down to 135 and back constantly, so often that overclocking actually gave me worse performance. After I flashed I got it to run at 993 mhz constantly, no weird throttling issues, no need for ThrottleStop, and my framerates are better than ever!

Keep in mind that, at a certain point, some games and benchmarks get bottlenecked at your CPU. One example is the "Ice Storm" test on 3Dmark. It's so easy for my 680m that no matter what I have the clock set to, and even if I have "prefer maximum performance" set, it'll still downclock to about 405 mhz before the usage goes over 50%. To get a higher score there I had to overclock my CPU, and even with it running at 4.2 ghz my GPU still stayed around 405. That's why they offer more stressful tests like Fire Strike; run those, not the dead-simple ones.

I'm not sure of any good programs that can monitor framerates, voltages or clockrates. I personally like HwInfo64 since it can send to Logitech keyboards with screens on them (Google Image Search "Logitech LCD"), but most people don't have those. The perky side is that your GPU doesn't need to draw anything extra on-screen, so as long as your CPU isn't the bottleneck you won't loose any performance.

HighBounce: Boot to DOS, run "nvflash --save MYVBIOS.rom", then reboot back to windows. Remember that DOS filenames can only handle eight characters, not counting the dot and three letters that come after it. You could use GPU-Z like slv7 said too, but if you run it in DOS instead then some of those pre-flash jitters you may have tend to go away.

slv7: So, I tried editing my vbios by hand. I have no idea how you do it. Apart from two error messages and the thing that appears when the system starts booting I can't recognize anything. My best guess would be something near 0001fe00, but I'm not sure what to change anything to (let alone how to adjust the signature afterwords!). So, if you have the free time, would you mind overvolting the Dell 5B 680m vbios for me? Maybe to 1.025, 1.050, 1.075 and 1.1? I posted my dump on page 179 but you also already have an unlocked edition on your second post, just nothing with an overvolt. Something that'd allow me to underclock it as well (since Optimus is completely disabled on this laptop; the monitor is connected directly to the GPU, not to my CPU) would be great. People say I "should" get seven hours of battery life, but I'm only getting three.

[Decoy101x]

well another problem stems from unreal engine 3 games when underutilize anyway. people from gtx titans to gtx660m's no matter the gpu will underutilize in those games.

[HighBounce]

Use GPU-Z (there's a button at the right side of the bios version line which will allow you to create a dump). Or use nvflash.

gpu-z tells me "bios reading not supported on this device"

how do i use nvflash to get the bios?

[drahcir_16]

Hello everyone. Newbie here, is there any stickies or link in the forum to get me started learning how to modify BIOS and vBIOS? I have Insyde BIOS from Dell Alienware. Thanks in advance.

[buluunee]

gpu-z tells me "bios reading not supported on this device"

how do i use nvflash to get the bios?

http://forum.techinferno.com/general-notebook-discussions/2166-%5Bguide%5D-nvidia-vbios-flashing.html read this one

[tigyi]

Just want to thank svl7 and any additional people responsible for these vBios files. The K5000m worked perfectly on my Dell M6600; Letting me raise the CPU clock from the locked 601 to 888mhz. memory +1000!

I should be able to go higher right? Nvidia Inspector 1.9.7 only gives me +135 base clock offset. I'd like to push 900 or 960mhz since at 888 the temps only get to about 76c.

Thanks again!

[Samelmamel]

Hi all. I have an MSI 16f3 with GTX680M and I always get lower than normal scores when overclocking.

Is the vBios able to fix this problem?

Sorry if this was mentioned already.

[svl7]

Just want to thank svl7 and any additional people responsible for these vBios files. The K5000m worked perfectly on my Dell M6600; Letting me raise the CPU clock from the locked 601 to 888mhz. memory +1000!

I should be able to go higher right? Nvidia Inspector 1.9.7 only gives me +135 base clock offset. I'd like to push 900 or 960mhz since at 888 the temps only get to about 76c.

You only get a max of +135MHz with my vbios? Which version exactly? And which driver are you using?

Hi all. I have an MSI 16f3 with GTX680M and I always get lower than normal scores when overclocking.

Is the vBios able to fix this problem?

Depends on what the issue is... what do you mean with 'lower than normal' scores?

[Samelmamel]

I was talking about 3dmark11.

People who overclocked their GTX680M got scores of around 7000. If i remember correctly that was at +135 core and +500 memory.

I used the exact same clocks but i only got around 6200 and i noticed that the graphic card was throttling during the benchmark.

So i'm just wondering if the vBios could fix this problem.

[svl7]

GPU score or total? My mods shouldn't throttle as long as your temps don't go crazy.

[Samelmamel]

Just ran 3dmark11 at +135 clock and +500 memory.

NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3610QM Processor, MS-16F3 score: P6253 3DMarks

Max temp was 70c.

[buluunee]

@Samelmamel there is nothing do with vbios or anything else. its EC failure. please refer to this thread : BETA EC for GT60 (16F3) & GT70 (1762)

edit: also throttling does not decrease 3d performance score so i would rather look for the problem somewhere around CPU? could you upload the 3dmark page which is yours?

[buluunee]

@svl7 here it is: GoGo Share and my current voltage which i see from gpu-z is 1.0370 that means i am sitting on V1030? or something else?

- - - Updated - - -

[ATTACH=CONFIG]6823[/ATTACH] here is the gpu z ss

@svl7 so i am sitting on 1030 max voltage or something else? coz seeing those gtx 680m vbios from the second post which are O,volted at 1050 or 1000 etc.

[Nofew]

buluunee: Throttling does decrease the 3D score. Did a benchmark before and after using slv7's vbios. Without it I throttle a lot when I have +135 set. The difference in the score's pretty big; 300-500 on Fire Strike.

In your second post, did you say you usually get 1.03 volts? Under what conditions? Is that constant or peak? Mine usually runs at .987 even when I have the clock set to 993, though for split seconds sometimes it goes to 1.012.

[buluunee]

buluunee: Throttling does decrease the 3D score. Did a benchmark before and after using slv7's vbios. Without it I throttle a lot when I have +135 set. The difference in the score's pretty big; 300-500 on Fire Strike.

In your second post, did you say you usually get 1.03 volts? Under what conditions? Is that constant or peak? Mine usually runs at .987, though for split seconds sometimes it goes to 1.012.

its at peak. and i did benchmark w.o throttle and throttle. it does not influence that lot. around 50 or something very small amount.

[Nofew]

Which test are you doing in which program, exactly? "Fire Strike"'s pretty intense; 500 means a /lot/. I really doubt it'd "lie" to me, so to speak.

If you're only seeing a difference of 50 it makes me wonder if you're doing "Ice Storm" or using 3Dmark 11. 11, on a 680, isn't all that accurate unless you're using the "extreme" preset. ("Performance" works too, sort of, but it can get bottle-necked by the CPU unless you have it at 4 ghz or so) -- The point is, if you're not using it already, go get 3Dmark (Just "3Dmark", that's their newest one), run Fire Strike and base your judgement on that. (You can't pick which tests to run unless you pay for it, but it does give individual results regardless. Scroll down and you should see it.)

Sort of off-topic but not. Sorta: Basically, 3Dmark 11's "extreme" preset doesn't lie to you. The score is how many frames you rendered, period. Nothing special. "Performance" multiplies by two (this can result in an odd number; remember that it counts partially-rendered frames), and "Entry" multiplies by four. Nope, kay, they do just adjust the difficulty. I think. A "perfect" score, not counting the physics or combined tests, is 7,200 frames. I figure the extra stuff must come from finishing those two faster than you "should" or something. I don't know. The rest of this post is still accurate, go hug it. (See what happens when I stay up an extra four hours?)

Alone, this isn't a problem. What causes the problem is that 3Dmark 11 has vsync enabled. In other words, if you /can/ render over 60 frames per second, it won't let you. There's basically a hard-coded limit for how well you're allowed to do, /and it does affect your score/. In the "Performance" preset, part of the first three tests usually hits 60. This can cause odd patterns in scores to come up. The reason so many people still use the Performance preset is because it's easy to get a "perfect" score, so to speak. If you hit 9,000, that generally means you can't do any better since you're at 60 FPS all the time in all the tests, and realistically no game is going to push your GPU harder (apart from Crysis and stuff like that, but those are /made/ to drive overclockers batty).

Even though you can get accurate results using Extreme (since that's definitely not going to hit 60 unless you've got a dual-GPU setup), the fact it's using vsync makes it iffy for me. "3Dmark" (not 11!) doesn't have vsync enabled. If you can pump out more than 60 frames on a certain test, it'll let you and your score will still be accurate. That being said, if you're getting over 1,000 FPS, there's a good chance that your CPU is the bottleneck and it's ability to "feed" the card is becoming a factor; make sure you run a test that suits your card. In general, if you have anything over a 660, you should be doing Fire Strike.

...I probably shouldn't be posting when I'm running off of no sleep, but eh. Here goes.

[andir]

Physics on 3d mark is computed by your CPU afaik .... So the problem is your CPU. Is it an older CPU?

Sent from my Nexus 4 using Tapatalk 2

i have an 3840QM CPU with 2.8 Ghz i dont find any thing to solve this problem with only 800 points...

NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3840QM,CLEVO P170EM score: P2874 3DMarks

here are my result

[tigyi]

You only get a max of +135MHz with my vbios? Which version exactly? And which driver are you using?

I'm using Nvidia Quadro Notebook Driver 311.35

vBios: 80.04.12.00.03 - 758MHz - 'OCedition'

(hopefully the attachment appears) You can see the slider maxes out at +135 and the Un "Relock Max" is grayed out.

I tried editing Nvidia Inspector's 'clocks shortcut', but it caused instability (I'm sure due to my inexperience in that area).

Is the +135 max due to the driver version I am using? All my other systems (with different cards and drivers) don't have such a low offset.

For the 311.35 drivers to properly install I had to edit the 'ListDevices.txt' and 'nvdmq.inf' files with the proper 'device ID'

History: I used the 311.35 laptop drivers first, then I tried the modified .inf and v314.22 drivers from 'LaptopVideo2Go' and something didn't/couldn't install; I think it was OpenCL? - without hesitation (or experimentation) I instantly went back to the self modified 311.35. This was just before I flashed your vBios; same driver remained installed after the flash.

So that +135 is due to the driver?.. I'll experiment with v314.22 again and report back shortly.

Thanks svl7 for taking an interest in the situation.

[svl7]

I think that that's an issue with my mod. I'll post something after dinner.

[buluunee]

Which test are you doing in which program, exactly? "Fire Strike"'s pretty intense; 500 means a /lot/. I really doubt it'd "lie" to me, so to speak.

If you're only seeing a difference of 50 it makes me wonder if you're doing "Ice Storm" or using 3Dmark 11. 11, on a 680, isn't all that accurate unless you're using the "extreme" preset. ("Performance" works too, sort of, but it can get bottle-necked by the CPU unless you have it at 4 ghz or so) -- The point is, if you're not using it already, go get 3Dmark (Just "3Dmark", that's their newest one), run Fire Strike and base your judgement on that. (You can't pick which tests to run unless you pay for it, but it does give individual results regardless. Scroll down and you should see it.)

Sort of off-topic but not. Sorta: Basically, 3Dmark 11's "extreme" preset doesn't lie to you. The score is how many frames you rendered, period. Nothing special. "Performance" multiplies by two (this can result in an odd number; remember that it counts partially-rendered frames), and "Entry" multiplies by four. Nope, kay, they do just adjust the difficulty. I think. A "perfect" score, not counting the physics or combined tests, is 7,200 frames. I figure the extra stuff must come from finishing those two faster than you "should" or something. I don't know. The rest of this post is still accurate, go hug it. (See what happens when I stay up an extra four hours?)

Alone, this isn't a problem. What causes the problem is that 3Dmark 11 has vsync enabled. In other words, if you /can/ render over 60 frames per second, it won't let you. There's basically a hard-coded limit for how well you're allowed to do, /and it does affect your score/. In the "Performance" preset, part of the first three tests usually hits 60. This can cause odd patterns in scores to come up. The reason so many people still use the Performance preset is because it's easy to get a "perfect" score, so to speak. If you hit 9,000, that generally means you can't do any better since you're at 60 FPS all the time in all the tests, and realistically no game is going to push your GPU harder (apart from Crysis and stuff like that, but those are /made/ to drive overclockers batty).

Even though you can get accurate results using Extreme (since that's definitely not going to hit 60 unless you've got a dual-GPU setup), the fact it's using vsync makes it iffy for me. "3Dmark" (not 11!) doesn't have vsync enabled. If you can pump out more than 60 frames on a certain test, it'll let you and your score will still be accurate. That being said, if you're getting over 1,000 FPS, there's a good chance that your CPU is the bottleneck and it's ability to "feed" the card is becoming a factor; make sure you run a test that suits your card. In general, if you have anything over a 660, you should be doing Fire Strike.

...I probably shouldn't be posting when I'm running off of no sleep, but eh. Here goes.

here i s min 3dmark 1.0 NVIDIA GeForce GTX 675MX video card benchmark result - Intel Core i7-3630QM Processor,Micro-Star International Co., Ltd. MS-1762. btw its higher than stock gtx 680m score and i got that score with throttling coz i am sitting on the stock voltage and higher the core higher the throttling as i examined. and also if your running on MSI machine then throttling causes are mostly related to EC.

[tigyi]

Alright, thanks svl7; looking forward to it

+ in it's current state, it performs MUCH better than the stock that came with the card (80.04.14.00.03) and FASTER defaults than the stock dell version I was using (80.04.33.00.35) - Why would it be a lower version number though? I'm assuming the version changes per distributor? Sorry for my ignorance on the subject, I'm a breadboard soldering/audio/graphics/2D-3D animator person... I dabble with web coding and would love to find a starting place in Software Engineering (specifically for hardware and web applications to compliment my other skills.)

-small note-

Just tried the 314.22 with the same results

I haven't investigated why OpenCL doesn't install with that driver version, I used the wizard, it might be user-error... not important, going back to 311.35 again.

[svl7]

@tigyi: K5000m version should be fixed now, check the second post. Let me know if all is fine. About the version... well, I only modifed that old one so far, the 80.04.33.00.35 is more recent, when I find time I add some more recent versions as well. Some people asked for overvolted versions anyway.

About the driver... no idea, I thought Quadro cards had different drivers for professional applications, but I might be mistaken.

[roo7y82]

hi svl7 hope you doing well, any news for 670mx medion bios ?

[tigyi]

About the driver... no idea, I thought Quadro cards had different drivers for professional applications, but I might be mistaken.

Yes, there are a few specialty drivers primarily for AutoCad or 3DS Max; I should have specified v311.35 ODE version; I apologize. With v314.22 I was attempting to substitute the GeForce drivers in its place.

Your changes worked!

My "Base Clock Offset" will now go higher (will experiment further later).

I didn't know if there was a difference other than distributor (dell, clevo..etc) when it came to vBios versions. Never knew such vast improvements could be made; thought it was determined by the driver. Wonder if a 'change history text' is available to see if there are any notable improvements between the different versions?

I might be too scared to try any of the overvolted versions (the M6600 only has cooling attached to one side of the card), but I am very much interested in any future developments.

Thanks again! Your work is very appreciated; I'll be sure to "buy you a beer" soon as I can.

[svl7]

hi svl7 hope you doing well, any news for 670mx medion bios ?

Oh right... damn. Check back a bit later tonight.

Your changes worked!

[ATTACH=CONFIG]6877[/ATTACH]

My "Base Clock Offset" will now go higher (will experiment further later).

I didn't know if there was a difference other than distributor (dell, clevo..etc) when it came to vBios versions. Never knew such vast improvements could be made; thought it was determined by the driver. Wonder if a 'change history text' is available to see if there are any notable improvements between the different versions?

Cool

As for the change log... hehehe, that's nvidia we're talking about, they won't tell us anything