[MOVED] Hacking NVidia Cards into their Professional Counterparts

[Garincho]

Hello, i have an Palit GT640. How to modify it to GRID K1?

[mooves]

Hi,
I did some research on my own about the GTX670/660ti vendor layouts because I wanted to have a cheap Grid K2 from ebay.
I want to share with you the things I found out the last two days.
Please be aware that this is only theoretical stuff, so I never tried it out by my own and it's on your own risk if you try it out!
If you want to try it, it would be kind if you reply if it works.

Now the explanation:
1. By comparing the various GTX670/660TI layouts I found out that the Gigabyte 660TI layout seems to be exactly the same layout like the normal original NVIDIA GTX670 layout.
2. I found a high resolution picture of the  Gigabyte 660TI back- and frontside and seperated the important views of  capacitor 1 to 4 (picture 1).
3. By random I saw that fortunately the capacitors have names beneath them (e.g. R137).
4. I seperated the capacitor names as you can see on the right side in picture 1.
5. I also compared the layouts of various vendors and found other vendors with the same layout like the Gigabyte 660TI. I wrote the models ontop each different layout. Please reply whether I made a failure.
6. By searching throught the various vendor layouts I recognized that the Asus 660TI Cards and also the MSI 660Ti Cards have named capacitors.
7. The Asus 660TI has the same layout like the Asus 670. By writing up the names of the capacitors again I found the probably right capacitors you have to modify (have a look at picture 2 and 3).

Here some overview:
NVIDIA GTX670 Layout:
Gigabyte GTX660TI
Zotac GTX660TI AMP!
PointOfView GTX670
Evga GTX670
NVIDIA GTX670

ASUS GTX660TI DirectCU II Layout:
ASUS GTX670 DirectCU II
ASUS GTX660TI DirectCU II

MSI GTX660TI TI Power OC Layout:
MSI GTX660TI TI Power OC

NVIDIA GTX680 Layout:
Gigabyte GTX670
Zotac GTX670
NVIDIA GTX680

I also uploaded a gif to compare the various cards more easy:
http://i.giphy.com/26FPxZopdlv3Vnqog.gif

Hi All,

Firstly I appreciate that this is a 4 year old forum but hope that someone can help.

I have an Asus GTX680-DC2-4GD5 DirectCUII Card which I have modded. I checked about 10+ times to make sure mine matched the Asus_Layout.jpg image posted. My first test was to get it to Tesla K10. I removed R4 as per Asus_Layout.jpg image and device ID went from 1180 to 11B0. My understanding is that removing R4 (R137) the Device ID should have changed the last (4th) byte to F. I have checked and there was no resistor in location R3 (R626) and certain I didn't remove that one too.

Have I misread the correlation between the table below and Asus_Layout.jpg?
I understood this to be: Resistor 0 is R1, Resistor 1 is R2, Resistor 2 is R3 and Resistor 3 is R4?

GPU Name   Resistor 0 / 3th byte   Resistor 1 / 3th byte   Resistor 2 / 8-f 4th byte   Resistor 3 / 0-7 4th byte
tesla k10   none   25k   40k   none
grid k2   40k   none   40k   none

Any help is much appreciated.

[gnif]

Hi All,

I stopped following this thread quite some time ago, my instructions were provided to help others identify how to locate and modify the other cards. For those that are new members/single posters that are just asking 'how'... my advice is:

• Read this thread from the beginning, we discovered some dead giveaways (SPI flash pinout) on how to identify which resistor contributes to which bits in the ID on this series of card.
• Simply removing a value is not a great idea, the input will "float" and the card can and will randomly change it's ID each boot, you must tie the input high or low with the correct resistor value for reliable operation. For those who have it working without doing so is just lucky, for most it wont, and for the GTX690, I can say for certain it does cause a problem.
• If your card has not been documented here already and you don't understand the terminology, or do not know what a resistor is, what it means to 'tie high or low' or 'float', or how damaging a multimeter on diode test is to the GPU you should avoid the attempt as the chances of destroying your card are very high.

If you understand all this and have the guts to experiment, here is how (today after learning much in this industry) I would go about testing the unknown resistors on other cards.

Needed Tools:

• Something to take photos with
• A decent image manipulation program (The Gimp is great for this)
• A decent multi-meter that does not exceed 1.0V on ohms test
• Kynar Wire (you can use PVC coated wire-wrap wire, but its not worth the hassle, use Kynar!)
  
• A variable resistor wheel or something similar

Method

• Photograph the front and back of the area you intend to probe, use a program like 'Gimp' to overlay the two images on each other with transparency and align them, this will make tracing the PCB MUCH easier!
• Use a multi-meter on ohms!!! I can not stress this enough, the diode test mode of the meter will dump up to 3V into the GPU which is a 1.2V part! It WILL destroy it!
• Identify which resistors are pulling up vs those pulling low, and mark them on your drawing.
• Tack a wire onto the common VCC rail (for pulling up), anywhere will do, just ensure it's the correct rail as the card usually has a few different voltages on it.
• Tack a wire onto the ground, I usually use a easy location to solder to for this, or even use an alligator clip on the edge connector of the card or computer case.
• Choose a un-populated resistor to test and tack a wire onto the side that connects to the GPU, not the side that is connected to VCC or GND.
• Attach your resistor wheel to one of the earlier wires for VCC or GND, the other to the wire you tacked onto the pad.
• Set the resistor wheel for about 10K, a nice safe value to start with. Anything lower then 1K is dangerous and could damage the GPU.
• Power up the PC and check the PCI ID, I found this is easiest done with the NVFLASH tool as it will report the IDs detected without the need to boot into a full operating system for each test.
• Write down the ID you got, compare it to the original if it changed and mark on your drawing/image what the pad did to the ID, or if it did nothing, mark that too so you don't lose track of where you are up to.
• Power off the PC, adjust the resistor value, and repeat the process

Eventually you may need to start removing resistors to test the other pads, the process is the same, just carefully remove the resistor first and then measure it and write down what it's value was.

Now if all this is too complicated for you and you don't have the time I am sorry there is little more I can do, unless people are willing to donate cards to reverse engineer and hack on I can not offer any additional help. If there is genuine interest in this I would be inclined to actually post a video of this entire process, but again this would require some sacrificial/donor hardware.

PS: Donor hardware does not have to be working properly, the only requirement is that the PC posts on it! If windows wont work with it, etc... it doesn't matter. The adjustment targets the PCI ID only, the card doesn't need to actually function properly to perform this work.

[protator]

Is there a reason still no one has tried to convert a 970 into a M4000 ?
Or a 960 into a Tesla M4 ?

I just blew my budget on a 2687WV4 and 128gigs so in order to complete my new WS I'll have to canibalize my old gaming rig, which so happens to have three 970s in it.
I've read somewhere that it's possible to SLI Quadros even without certification.
Three Franken-Quadros and a couple ebay-960s turned M4 could turn this into a really sweet machine.
+++make SolidWorks fly me wants+++

[protator]

Hmmm, threat seems dead. Too bad.
Guess I'll have to go ebay hunting for a few sacrifices and try this on my own.

[protator]

Does anyone here know where I can find a bios file for a M4000? Or the M4.
Couldn't find it in the Techpowerup database, unfortunately. 
Only Keppler cards list there.

[gnif]

Hmmm, threat seems dead. Too bad.
Guess I'll have to go ebay hunting for a few sacrifices and try this on my own.

Sorry mate, it mostly is. The biggest issue is, that nobody has to my knowledge tried this with the newer maxwell or pastel cards, either new to cost or the fact that Maxwell introduced signed bioses, etc. I bet that altering the card's behavior is likely much more involved now.

[whiskthecat]

Someone had asked about converting 780ti dual bios EVGA classified. I can confirm that the 18K resistor from Vcc to SCLK on one of the 2 EEPROMS works as expected and you can even use the selector switch to change between geforce and quadro.

[fixator]

Not sure if anyone is still following this topic, but I would like to request some help identifying R1-R4 on this card. The card is an Asus GT640-DCSL-2GD3. I have posted pictures below showing U10 and the surrounding resistors. Please let me know if you need any more information!

[variance]

Hi everyone.. I have a 04G-P4-2647-KR (EVGA GT640 4GB) trying to convert to a Grid K1.
I think I've found the straps on the front of the board near the bios.

Could someone confirm if this looks right.. or confirm that these are the correct resistors? 
According to the table in one of the OPs. The GT640 should have 25K at R1 and 10K at R3.

I think I've identified R1 and R2 but have two possibilities for R3 and R4.
either R3 and R4 are after skipping one resistor and directly above R1/R2 as both those test at 10k.
or R3/R4 are across and down to the right as that's the only other one that tests at 10K.

Can I get confirmation as to which one is more likely/logical?



If I've read everything correct. In order to get the Grid K1 ID on a GT640
 #1 I need to remove R1 resistor.
 #2 Solder 40k SMD resistor to R2 spot
 #3 Remove R3 10k resistor, Solder 15k Resistor to that spot.

Could I get confirmation on that methoology and also a recommendation on the SMD resistors to get.
like tolerance spec or brand?

High-res pic of entire board here:
http://imgur.com/a/khrKt



 

[gnif]

Variance, the circuit for each strap is simple, it's either a pull up or pull down for each nibble.

Let me make this clear as many people have missed this, there are two possible resistors for each nibble, one pulls up, one pulls down, only one from factory will be populated, the other unpopulated.

1) Use ohms on you meter to figure out which pads are ground, take a photo and note them all (I just put a colored dot on the photo)
2) Use ohms on your meter between the VCC pin of the EEPROM and the pads to figure out which are VCC and note them all (again, just a different color dot).
3) Now you need to identify which positions are common for pull up/down, so again on ohms, measure between the pads that are still unknown looking for two that are connected.

At this point you will have a good idea of which resistors are worth looking at. To discover which is which, since the nibble is changed every 5K, solder a 10K resistor over the already existing 10K resistor to test, this should change one nibble as it will drop the resistance to 5K. If it doesn't work, its not the right resistor.

Once you have identified the resistor for each nibble, and know which set of pads are for the opposite (pull up/down), you can then figure out what needs changing. Some resistors may need swapping with different values, depends on your target ID. Others may need removing and populating the other set of pads for that nibble to pull it in the opposite direction.

From factory none of these will be floating, you should not have more or less resistors on the PCB after you have done your mod, if you have, you have done it wrong.

Leaving a resistor off will cause the input to float, and the nibble ID will be undefined, it could change from boot to boot causing you odd problems.

[variance]

Variance, the circuit for each strap is simple, it's either a pull up or pull down for each nibble.

Let me make this clear as many people have missed this, there are two possible resistors for each nibble, one pulls up, one pulls down, only one from factory will be populated, the other unpopulated.

First of all, Thank you for your patience and willingness to explain in detail. (Incidentally Is there a sticky somewhere I'm missing for a guide on terminology like "nibble" or "strap"?)
Secondly,
From the first few dozen posts scattered. I've seen that there are 4 spots that are referred to as
R1, R2, R3, R4. It's frequently indicated that only two of these 4 are populated.
Am I to infer that R1, R2 would be pull up, or pull down on the value depending on the resistor and that R1/2 are the common pads for one nibble and R3/4 are the common pads for the other?
Looking for clarification on this point.
I hope to create a better guide/post with pictures after I successfully complete the mod a few times.

1) Use ohms on you meter to figure out which pads are ground, take a photo and note them all (I just put a colored dot on the photo)
2) Use ohms on your meter between the VCC pin of the EEPROM and the pads to figure out which are VCC and note them all (again, just a different color dot).
3) Now you need to identify which positions are common for pull up/down, so again on ohms, measure between the pads that are still unknown looking for two that are connected.

Could you explain the methodology or point to resources on this topic? So far I've been just placing my multimeter in OHMs mode
and placing the probes at either end of the resistor. I have some assumptions on the steps but I'm not sure if they'd be correct.

1. Identifying ground/vcc: Would I be correct in connecting my black test probe to the ground lead on the PCI-E slot and the red probe to either end of the resistor until I get the known value and then assuming that end of the resistor is the VCC line and the other one is the ground?

2. VCC pin on the EEPROM.. would this be pin 3? not entirely sure what the protocol for doing this is either.

and on another topic... what's the newegg or amazon equivalent people here use to order components like smd resistors?

[gnif]

You're welcome . No sticky that I am aware of, a strap is a hardware configuration input that is used to perform very very early setup, before the chip even starts to load from the EEPROM. In this instance tells the GPU what device it should be. I have a suspicion that enabling the extra Quadro features is just another strap that has not been discovered yet, but I am not willing to probe this card more until I replace it with something new (not too long now).

R1, R2, R3, R4. It's frequently indicated that only two of these 4 are populated.

You are correct here, just tried to be extremely plain and simple in my information as some people have tried adding the extra resistors without first removing it's opposite, forming a resistor divider and not seeing the truth table in the first post I made hold true.

So far I've been just placing my multimeter in OHMs mode and placing the probes at either end of the resistor. I have some assumptions on the steps but I'm not sure if they'd be correct.

Perfect, that's what you want, the reason I stated this is because many novice or general 3.3V-5V engineers would just without thinking default to using the meter on diode test to buzz out the circuit, but since the GPU runs at barely over a volt, the diode test on most DMMs would damage the GPU as you probe around. The straps seem to be 3.3V tolerant as that is what the EEPROM runs at, but touch a trace that isn't for the EEPROM and you might kill the GPU.

1. Identifying ground/vcc: Would I be correct in connecting my black test probe to the ground lead on the PCI-E slot and the red probe to either end of the resistor until I get the known value and then assuming that end of the resistor is the VCC line and the other one is the ground?

Yes, but make sure the negative probe is on the chassis/ground, so any induced voltage doesn't damage anything. Find the ground plane first and then test across the resistors from the chassis/ground point, to the resistors and find the common ground pads. When looking for VCC, do the same, positive on the known positive point, again so as to not risk reverse polarity on the parts.

 

2. VCC pin on the EEPROM.. would this be pin 3? not entirely sure what the protocol for doing this is either.

Not sure, you need to lookup the part number (google it) and find it's datasheet, it's likely a jellybean SPI flash chip, I do not have enough experience though to confirm if the pin out on these parts is all the same across all brands (seems to be from what I have seen, I think its part of a JEDEC standard).

and on another topic... what's the newegg or amazon equivalent people here use to order components like smd resistors?

I used DigiKey, but for the purposes of testing I used some kynar wire (aka, wirewrap) so I wasn't constantly soldering on the PCB and swapped out 1/4W 1% cheap resistors till I found the pattern and the values to use. After this I ordered the exact values I needed.

[mttg]

Hi All,

It is possible to make a GRID K1 from GT640-1GD5-L?

mTTg

[mttg]

Nobody a idea? 

[gnif]

Read the thread through, figure it out, all the information is there.

[mitac]

Hello
I have a GeForce GTX 980 4GD5 OCV1. Is it possible to modify Isa larger quantities streams NVENC. Currently, card transcodes me only 2 channels. Unfortunately I never found the instructions to change resistors.

Regards

[gnif]

I stopped following this thread quite some time ago, my instructions were provided to help others identify how to locate and modify the other cards. For those that are new members/single posters that are just asking 'how'... my advice is:

Read this thread from the beginning, we discovered some dead giveaways (SPI flash pinout) on how to identify which resistor contributes to which bits in the ID on this series of card.
Simply removing a value is not a great idea, the input will "float" and the card can and will randomly change it's ID each boot, you must tie the input high or low with the correct resistor value for reliable operation. For those who have it working without doing so is just lucky, for most it wont, and for the GTX690, I can say for certain it does cause a problem.
If your card has not been documented here already and you don't understand the terminology, or do not know what a resistor is, what it means to 'tie high or low' or 'float', or how damaging a multimeter on diode test is to the GPU you should avoid the attempt as the chances of destroying your card are very high.

[Duke711]

sorry for my bad englisch. I hope you understand me. I would like to modding GTX Titan 6GB to a Quadro 6000. I should to install a 18k resistor, by the following instruction:

On a separate note, I just learned that GTX780Ti has device ID 0x100A. K6000 is 0x103A. We only need to change the 3rd nibble (via oguz286's awesome mod).
His mod is to use a 33K resistor to jack up the 3rd nibble from 0 to 2. On my Titan using an 18K resistor instead boosts the ID to 3. So to make a 780Ti to a K6000,
simply apply an 18K resistor between VCC and SCLK and voila, job done. For extra points, solder a couple of wires between those pins, solder an 18K resistor on one of them,
and a switch to connect them. Break the switch out somewhere accessible (extra extra points for making the switch easily and neatly accessible from the back of the card without
obstructing the airflow too badly). Now you can switch betwee a 780Ti and a K6000 at a flip of a single switch.

I have not quite understood where I have to place the 18k resistor. Could someone explaining me on the basis of the following graphic, please?

http://vps1931.directvps.nl/GTX780-1.jpg

At 0, 0.3, on the left of 25k, right of 45k, or a resistor must be to replace?

Thank you

[gnif]

I have not quite understood where I have to place the 18k resistor. Could someone explaining me on the basis of the following graphic, please?

I stopped following this thread quite some time ago, my instructions were provided to help others identify how to locate and modify the other cards. For those that are new members/single posters that are just asking 'how'... my advice is:

Read this thread from the beginning, we discovered some dead giveaways (SPI flash pinout) on how to identify which resistor contributes to which bits in the ID on this series of card.
Simply removing a value is not a great idea, the input will "float" and the card can and will randomly change it's ID each boot, you must tie the input high or low with the correct resistor value for reliable operation. For those who have it working without doing so is just lucky, for most it wont, and for the GTX690, I can say for certain it does cause a problem.
If your card has not been documented here already and you don't understand the terminology, or do not know what a resistor is, what it means to 'tie high or low' or 'float', or how damaging a multimeter on diode test is to the GPU you should avoid the attempt as the chances of destroying your card are very high.

[variance]

ISA larger quantie stream NVENC?
If the # of streams is limited by driver and would work with a Quadro or Tesla. than maybe.
The professional cards that share common core (GM204) with your GPU are the:  Quadro M4000/M5000, Tesla M6, Tesla M60
but from what I've gathered reading the entire thread and preparing to mod my keplers. (2 GT640s and 1 GTX690).

1. I don't believe anyone has attempted (That I'm aware of) mod a Maxwell GPU in this manner.
2. The Leg work hasn't been done/documented to identify whether or hardware strap ID is dealt with identically to the kepler
    nor has anyone done the leg work to identify what the hardware IDs and what the resistor values are between the GTX, Quadro and Teslas
    for maxwell.

Hello
I have a GeForce GTX 980 4GD5 OCV1. Is it possible to modify Isa larger quantities streams NVENC. Currently, card transcodes me only 2 channels. Unfortunately I never found the instructions to change resistors.

Regards

[coolnik]

Hello! Tell me which resistors to replace this video card asus GT640-2GD3. Photos of the video card below.

[gnif]

Hello! Tell me which resistors to replace this video card asus GT640-2GD3. Photos of the video card below.

Umm no... read the thread, figure it out, each card is different and we do not have the resources to obtain each one to find them.

[coolnik]

Well, after reading about the Gt640 is written about the zotak and the asus of the other is not more than me. And there is another list that managed to successfully model I want to understand what to buy? We have a small choice to find ...

[variance]

Hi. I just modded a zotac GT640 into a Grid K1...
It identifies as a Grid K1 under nvflash but under esxi shows up as a Quadro 410.
Has anyone experienced this issue?