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

[svl7]

Is it the 4GB or the 2GB version?

[Muerte]

I think come old 2GB version.Will be Dell refurbished.

Trying ask 4GB version? Come 2 piece,my laptop M17xR2,my friend XMG P701

[svl7]

The 4GB version is the same as the one Clevo uses. The 2GB is slightly different, but I think it should work fine. Possible issue could be audio per HDMI in the Clevo system.

[Muerte]

Understand,many thanks.I not yet read totally this topic.Which version better? I think temp,vbios working issues.

[Kyung Ho Sin]

My laptop has gtx 675mx and I'm using vbios. I'll change my graphic card to 680m, then what should I do?

[buluunee]

My laptop has gtx 675mx and I'm using vbios. I'll change my graphic card to 680m, then what should I do?

you will regret if you upgrade your gpu to 680m coz you will see a small amount of improvement. I would rather w8 for 780m card to be released. Even i have 675mx and ive seen what 680m can do and decided to w8 for 7xx series.

[Brian]

My laptop has gtx 675mx and I'm using vbios. I'll change my graphic card to 680m, then what should I do?

Nothing, it should be plug and play as long as your P150em bios supports it.

[Brian]

you will regret if you upgrade your gpu to 680m coz you will see a small amount of improvement. I would rather w8 for 780m card to be released. Even i have 675mx and ive seen what 680m can do and decided to w8 for 7xx series.

It probably isn't worth the performance difference going from 675MX to 680M. I agree he'd be better off waiting on the 780M which should have the same amt of cuda cores as the 680MX.

[Khenglish]

It probably isn't worth the performance difference going from 675MX to 680M. I agree he'd be better off waiting on the 780M which should have the same amt of cuda cores as the 680MX.

I agree. The performance gain is not worth the $600+ cost of the 680m. I suggest waiting for the 780m and either getting that or getting the 680m then since it will be far cheaper.

[majm84]

Hi,

could somebody please modify Fujitsu-Siemens NH532 bios so I can get past 135MHz limit? Here is the link for the latest version 2.07:

https://docs.google.com/file/d/0B3Q-OQTLebHLWWpoWVZIekdlR3M/edit?usp=sharing

[Nofew]

The limit is on your GPU, not your BIOS. Please provide your vBIOS instead. There's instructions over yonder . Just follow them until you're in DOS, then do "nvflash --save myvbios.rom" and toss that up here, along with any information you can (Series, amount of RAM, vendor (of the GPU, not your entire computer! This won't be "nVidia".), etc).

Note that the forum doesn't accept .rom files. You'll need to zip it in order to upload it.

[svl7]

No... I need the bios, just as majm84 posted, that system has an 640m LE which is soldered to the board.

[Nofew]

Suddenly, silence. Guess that makes it okayish for me to just ask..

Is there any chance you can mod the Dell 2 gig 80.04.5B.00.02 vbios to have some extra voltage? I've asked before but you seem to have skipped over my posts. Maybe they were too long.

If it's not too much trouble, could you also raise (or lower) the minimum clock limit so I can bring it down to like, 10 mhz? I like seeing what happens when I underclock stuff.

[Trach62]

Hi

i just O/C with modded bios and got this NVIDIA GeForce GTX 675MX video card benchmark result - Intel Core i7-3630QM Processor,Micro-Star International Co., Ltd. MS-1762 score: P5837 3DMarks

but i would like to modify the voltage, how can i do ?

[Nofew]

Trach62: You can't. Apparently the voltage modifier in Afterburner and such can't be unlocked (or it can be but svl7 hasn't found the right things to play with), so instead you have to ask for a vbios that comes with a pre-modified voltage.

Maybe if we get enough people together asking for overvolts he'll do them all at once. That'd be shiny! And zappy.

[Trach62]

So people who scored P6k5 on 3Dmark11, it's because they have a vbios already overvolted ?

[schranzmaster]

Maybe if we get enough people together asking for overvolts he'll do them all at once. That'd be shiny! And zappy.

i would like a overvolt for my gtx 660m

[Nofew]

Trach62: No. There's better cards than what you have, like mine. I typically get P7.7k with no overvolting.

"Mobile" nVidia cards are pretty much just desktop cards with a vBIOS that limits overclocking. They're otherwise pretty much identical. The 675mx is a desktop 660, the 680m is a desktop 670, and the 680mx is a desktop 680. Notice how the core count and memory transfer rates match, even though the clockrate and memory capacity differ. (Remember that we're just comparing the GPU core, not the entire platform. The core is the only part nVidia makes; the rest comes down to Dell, MSI, Galaxy and whoever else decides to purchase from nVidia.)

[Trach62]

i meant that people scoring 6k5 with gtx 675mx have vbios overvolted !

i know there is better card than mine ^^

[harryob]

i meant that people scoring 6k5 with gtx 675mx have vbios overvolted !

i know there is better card than mine ^^

I've been able to get around 6500 with my 675mx and no overvolt, but it's not stable. To do it I overclocked +405 core, +800 mem. 3d Mark will somtimes give a result of around 6500 with that overclock, but if I run Heaven or another very taxing program the driver will crash. I've found that max stable for my card is around +395 core, +600 mem. I reduce that to +385/+500 for daily operation and still get a 3D Mark of 6200-6300.

[Trach62]

when i go +350 core it is ok but when i go 351 , 3D mark crash ! same for the memory , when i reach 400 => 3D mark crash :/

[Nofew]

You might just have a bad chip, then. Technical stuff below! You don't need to read it. Long story short, some chips just don't come out well and can't be overclocked as far.

When microchip manufactures make their devices, they don't actually know what'll come out until it's done. Let's say you're Intel. You have three chip designs you can make today: i3, i5 and i7. Let's say you settle on i7's for the day. You get the mold (which is basically a film like you'd find in a camera) and shine a light through it onto a large wafer. Anything that's not hit by that light will then be etched away by dumping it in a tank full of corrosive chemicals. Once it's done, each copy is cut out (and there's hundreds to thousands per wafer) and some further assembly is performed. At the end of all that, each chip is tested by a "chip tester" that essentially asks them very low-level questions and makes sure they all come out right.

Now, here's the catch: About half will be wrong on the entire chip that it's just discarded. Of that half that makes it through, not all of them are perfect! Did you know that every i7-3930k actually has /eight/ cores, which actually makes it a Xeon, but two are defective and are thus deactivated in firmware and then sold as an i7? Yeah. That's really what happens.

All chip manufacturers do the same thing. If they didn't, they'd be flat broke. GPU manufacturers are even more sensitive to these issues. The 680 GTX has 1,536 cores. That's a lot of places where things can go wrong and get it demoted to a 670 GTX. They hope for the best; they hope for all 1,536 cores to work just fine and the chip to conduct electricity as it should, they hope they can make 680's, but in general this won't happen. 660 and 670's are far more likely. It'd be stupid to throw out a partially broken product if it can still be fixed to work within lower expectations. I mean, if you have a bag of potato chips, you don't only eat the big whole ones and throw the rest away, do you?

After that, they still don't know what they've made. They can classify it as an i7, Xeon, etc, but don't know the model number yet. To find this out they put the chip on another device that carefully raises the voltage and clock until it finds a good balance. This is why some chips with the same model number use more or less voltage and are more or less overclockable.

A "golden sample" is like the Holy Grail for overclocking. Those are chips that have exceptionally low voltage requirements for their model number. Typically this happens in one of two ways: Either Intel ends up with an i7-3970x and says "We have too many. They're not selling. Let's put an i7-3930k label on it instead and sell it as that so whoever gets it has longer longevity and we don't have a surplus of stock.", or the sample scores just below the requirement to be an i7-3970x and becomes an i7-3930k. In either case, you suddenly have the power of an i7-3970x (except with only 12 megs of L3 cache rather than 15) with an i7-3930k label. You can overclock it as far as you would an i7-3970x.

You might be asking, "But Nofew, pray tell, if the factory sets the voltage the chip runs at, then shouldn't people with a Golden Sample still have the same overclocking potential as the rest of us since their vbios probably lowers their input voltage too?". You'd be right, that would be the case, if it wasn't for a guy named Werner Heisenberg and the fact circuits on chips are so small, under 30 nanometers.

There's this thing called the "Heisenberg Uncertainty Principle". (EDIT: Okay, apparently this isn't the correct name. What's coming next is all true, save for the name of the phenomenon.) Basically, it boils down to the fact that we have circuits so small that only about 100-200 electrons can actually fit in the circuits at a time. Voltage is the "speed" of the electron flow. The lower it is, the slower things will move and the less heat it will make. Amperage is how densely packed it is. Believe it or not, your processor uses over 40 amps. That's huge for small electronics. It's like trying to get a massive, constantly-moving group of people to walk across a bridge that's only 60 feet wide. Everyone has to stand shoulder to shoulder, and each row has 40 people in it. They have to move slowly or some will fall off, and they all need to keep moving so 40 come out at the other side every second. If less people stand in the same row, they need to run faster. If they run too fast the bridge will break from swaying too much, and this is the typical limit people run into with overclocking. However, when you're at this small of a scale, the fear is that people will fall off and you'll need to speed up more to keep the same throughput, which in turn knocks off more people and creates a nice feedback loop that inflates the voltage required.

Things are happening on such a small scale that errors can't reliably be detected. Higher heat and voltages cause these errors to happen more often. In other words, less is more! Seriously! There comes a point that decreasing the voltage is more beneficial than increasing it. Due to the way the atoms are acting, pumping more voltage in throws more out of whack than you put in and you're effectively pushing less voltage into the chip. Trying to cram less inside in the first place keeps things tame and lets them do their job properly. This is why Golden Samples are golden. They're below a critical threshold that lets their owners actually do this to them. Regular samples already have their voltages too high to work this way; they're going to keep leaking electrons too fast and there's nothing special that can be done.

That being said, it does not mean that if you have a Golden Sample that you always decrease the voltage rather than increase it. You still bring it up from time to time but there's different rules in play. The potential is higher, but there are certain "magic islands" that are kind of randomly left in the middle that you need to find rather than caring about your temperatures (since your voltages, if you find an island, will be lower than someone with a regular sample trying to hit the same clockrate). These islands are special little places where the Heisenberg Uncertainty Principle is kept at bay so the extra voltage is actually helping rather than hindering. They'll push the chip faster than non-Golden Samples at the same voltage can ever go. Finding them is complete guesswork in most cases; It's called the "Uncertainty Principle" for a reason!

Much like the Golden Samples, there are what I call "bad apples". These can come about in various ways, but in general they arise either due to a shortage of stock that forces the company to accept lower-grade components as higher-grade ones, they're damaged at some point during installation, or temperature kills part of them. In any case, these parts require extra voltage for some reason. This lowers their overclocking potential. These samples still follow the same rules as normal chips do when it comes to overclocking; they simply have less potential.

So yeah.

By the way, nVidia GPU's work in groups of 13 mhz at a time. Although you think you changed only 1 mhz, you really tripped over a line that caused it to change 13. Likewise, if you ran at +349, it'd be the exact same speed as +350. Same with 348, 347, 346 and so on, down to 338. Just sayin'.

[tigyi]

@Nofew

Great Article!

[Trach62]

+1 thanks for your explanation !

[Brian]

@Nofew very good post with a nice simplified explanation. Thanks.