[MOVED] Hacking NVidia Cards into their Professional Counterparts

[angerthosenear]

Hi guys!

I'm writing justo to confirm that everything worked as expected with my evga gtx680!!! Nos it is modded to a grid k2!!!


Thanks a lot for your great feedback!!!

Which resistors did you change? Did you have to swap out two resistors (which is what I'm thinking) or do something else?

I'm planning to do the same thing, but with a 690 (of which I only see it converted to Quadro or Tesla). I don't mind doing the extra solder work for the Grid card.

Thanks for any input.

Hi angerthosenear,

I've made this changes:

resistor 0: solder new 40k resistor
resistor 1: removed it
resistor 2: solder new 40k resistor
resistor 3: removed it

I'm experiencing some problems with passthrough mode in vmware view 5.3... I added k2 hardware in passthrough mode and then added that hardware to a virtual machine (as external pci). Windows XP testing machine detected it, but once I halted virtual machin to change network configuration, vmware server took sooo long to shutdown that virtual machine and had many stability problems. Once it reacted, passthrough hardware disappeared, and also the nvidia k2 dissapeared from hardware inventory in the host.

Any ideas guys?

Regards,
villa

Is resistor 1,2,3,4 from top to bottom as in the main post (in picture 2)? Or some other picture in this thread?

Thanks for the response. Not sure about your current issue however, I'm not versed in VM stuff at all so can't help there.

[dodgygeeza]

Hey everyone, has anyone here attempted to mod the EVGA GT640 4GB? I've been reading the thread about how the GT640 can be converted into a Grid K1 (which would be great to use with VMware), but from looking at the pictures people have posted it seems as if the board layout is somewhat different. I've removed the heatsink to get a full view of the board, but I still can't locate where the correct resistors are...

Has anyone here tried with this card before? or does anyone here have any idea where the correct resistors might be? I'd really like to get this card converted given its perfect for a 1u sever that is tight in space, and the 4GB RAM means it should play nice with vSGA as well :-) I can post some high-res pictures online if people want to take a look.

[djthrottleboi]

Has anyone attempted to mod a 780 into a titan? i want to see if i can get fp64. I want to use it in windows though. also my linux distro is solydk 64bit.

[villalvilla]

Hi guys!

I'm writing justo to confirm that everything worked as expected with my evga gtx680!!! Nos it is modded to a grid k2!!!


Thanks a lot for your great feedback!!!

Which resistors did you change? Did you have to swap out two resistors (which is what I'm thinking) or do something else?

I'm planning to do the same thing, but with a 690 (of which I only see it converted to Quadro or Tesla). I don't mind doing the extra solder work for the Grid card.

Thanks for any input.

Hi angerthosenear,

I've made this changes:

resistor 0: solder new 40k resistor
resistor 1: removed it
resistor 2: solder new 40k resistor
resistor 3: removed it

I'm experiencing some problems with passthrough mode in vmware view 5.3... I added k2 hardware in passthrough mode and then added that hardware to a virtual machine (as external pci). Windows XP testing machine detected it, but once I halted virtual machin to change network configuration, vmware server took sooo long to shutdown that virtual machine and had many stability problems. Once it reacted, passthrough hardware disappeared, and also the nvidia k2 dissapeared from hardware inventory in the host.

Any ideas guys?

Regards,
villa

Is resistor 1,2,3,4 from top to bottom as in the main post (in picture 2)? Or some other picture in this thread?

Thanks for the response. Not sure about your current issue however, I'm not versed in VM stuff at all so can't help there.

Hi Anger,

The photo you are asking for is this:

https://www.eevblog.com/forum/projects/hacking-nvidia-cards-into-their-professional-counterparts/?action=dlattach;attach=43392;image

It is in this forum on page number 6. You can also find the description from the first page on this forum, where the author talks about GTX680 already modded, on the first post.

Hope it helps,
Villa

[Allso]

I keep reading in this thread that there is no advantage to quadro over tesla yet I cant seem to find any info anywhere whether or not a tesla card will allow me to use 10 bit color like a quadro card. Perhaps some of you experts can tell me if this is the case or not, so I dont do a useless hack?

[vacaloca]

I was successful in soft-modding a Fermi GT440 3GB (OEM) HP part # 631078-001 that I had bought a while back -- http://www.ebay.com/itm/371054989914 into a Quadro 2000. I used the tutorial here: http://www.altechnative.net/2013/11/25/virtualized-gaming-nvidia-cards-part-3-how-to-modify-a-fermi-based-geforce-into-a-quadro-geforce-gts450gtx470gtx480-to-quadro-200050006000/

Here is a screenshot of the card running a PrimeGrid task on the GT440 (OEM) soft-modded into a Quadro 2000 running on a Windows 7 x64 VM:

You can see I also enabled TCC mode via nvidia-smi on next reboot. Not shown here, but TCC mode works, and PrimeGrid continues the computations.

I was also able to soft-mod a PNY GT430 into a Quadro 600, but later realized that:
1) TCC mode isn't supported
2) Quadro 600 isn't multi-OS capable, so the same code 43 error pops up in the Win 7 x64 VM.
So, basically, that mod was useless, so I undid it.

My platform is a Gigabyte X58A-UD3R v2.0 motherboard with Fh1 BIOS version (VT-d enabled in BIOS via Alt+F1) with an Intel Core-i7 930 CPU with ESXi 5.5 U1 (patched with ACHI & NIC drivers). I was also successful in doing passthrough with an AMD Radeon HD 6770 card with no modifications required on the same system.

[choPer]

greetings to all
my problem a little bit specific but maybe someone help me ?
i have a gtx770m card from dell with 1028 05AA vendor id and it's not working in my laptop (doesn't post) but gtx 770m clevo version (with the same pcb layout and 10DE 1008 vendor id) working perfect. Flashing bios from clevo didn't help too.
any ideas ?

[SoNic67]

I just registered here to say thank you to everyone that posted modding information on this thread.
I just soft-modded my GTS 450 into Quadro 2000 - now is basically a higher clocked Quadro. Changed the straps and the name of the card in the BIOS to say Quadro 2000 (for tidiness).
nvflash --index=X --straps 0x7FFC2FFF 0x10006000 0x7fffffff 0x00000000
After that I saved the BIOS with GPU-Z, correct the name with a hex editor and corrected the checksum with NiBiTor. Flashed back with nvflash.

I tested it at work with Cadalyst C2017 and final 3D result was 804 - while a Quadro 600 gets 705 and a Quadro K2000 gets 828 (those two are "real" cards that I could test). Don't have a real Quadro 2000 around to compare.

[slis]

Hello there,

i have read whole tread but will do it again... too much info )

Just few questions...

What is best gpu to modify ( i mean with smallest amount of modifications as possible ) to GRID K2? what manufacturer? with 4gb?
evga 680gtx 4gb?

Does it work as vGPU so 2 - 4 vm-s can share it?


Would it be possible to run couple of xen virtual machines with steam and games on it, then stream it using steam to low powered clients aka laptops with steam?

Thanks

[gordan]

Easiest to modify? GTX 4xx cards - they can be modified using the software strap in the BIOS, e.g. GTX480 -> Quadro 6000. This is VSGA capable.

For Kepler class cards, the easiest are:
680 - remove 1 resistor off the back off the card to make it into a Tesla K10. Don't modify it into a Grid K2 if you plan to use the video output on the card, it will only work in SL-DVI mode, unless you get an updated Grid K2 BIOS, modify it to add output port initialization, and even then it only works with very recent drivers (see discussion on this recently). K10 just works with no further modifications.
780Ti - Add an 18K resistor on the EEPROM pins, as discussed more recently on the thread - big resistor and easy to solder. This turns it into a Quadro K6000

All of the above should work just fine with VSGA.

[slis]

thx for quick and nice answer

i am xen noob... sry
but what is VSGA ? so i can share 1gpu to 2-4 vm-s? with full
3d support?

i thought only vgpu does that?

aka one 680 with 4gb as grid k2 -> 2vgpu with 2gb to each vm

edit:

i just searched about vsga, you mean vmware? its only dx9 i need 11


and i am ok with no output on gpu cause i will stream vm output with steam over network just for windows games... to linux clients

[gordan]

I'm a Xen user, too, but I use straight PCI passthrough. GPU sharing is a relatively recent project on Xen, and it is not mature and stable AFAIK. VSGA is VMware's GPU sharing, and that has been around for a while. VSGA works with Quadro 5000, 6000, K5000, K6000, Tesla K10, and Grid K1 and K2. I don't know what  the requirements are for Xen's GPU sharing.

[slis]

this is without passthrough in xen... u just have 3 modes for gpu 512mb, 1gb or 2gb frame buffer with full dx11 and it requries GRID K2 / K1... and u have nvidia drivers in client machine not vmware....

anyway back on topic..

i read that one guy just poped off 2 resistors off gigabyte 680gtx 2gb to get grid k2?  is that the simple one and is it the same for 4gb? i want to avoid soldering and have 4gb grid k2 

[gordan]

Yes, popping off two resistors worked for me without replacing them. Other people found they had to replace them to make it work. It varies. I use straight PCI passthrough for my multi-seat gaming/workstation Xen system (3 GPUs), so cannot really offer any advice on shared GPU solutions.

[slis]

Ok thank you again for your answers.

I'll go with evga or gigabyte 680 4gb so we will see how that goes 

[SoNic67]

Nobody has yet been able to restore the missing GL primitives on GeForce cards. Cross-flashing the BIOS, in the one case where it actually works (Q2000 BIOS onto a GTS450), doesn't seem to achieve anything obviously useful in this regard; then again, Q2000 and GTS450 are very, very similar, much more so than other GeForces are to their equivalent Quadros, with maybe the exception of a K5000/GTX680 4GB variants - in that they have the same amount of VRAM. I haven't tried flashing a full strap-adjusted K5000 BIOS onto my GTX680 yet - it is on my ever-growing TODO list.

In case of the GF106 GPUs (GTS450/Q2000) I suspect the missing functionality is cut out of the GPUs before packaging, and if that is the case, the chances of restoring this are non-existant.

Note, however, that modifying a GTS450 into a Quadro 2000 does produce some performance benefits - Maya scores, although still far behind a real Quadro 2000, go up by around 40% after modifying the card.

I just did mod my GTS450 into Q2000 and I am curious if you know what OpenGL primitives are missing. I can compare it eventually with "real" Quadro 600 or Quadro K2000.
However, note that newer AutoCAD and Revit software, on Windows 7&8 (x64), use DirectX11 to process the 2D and 3D, not OpenGL.

[thunderbolt]

Hi,

sorry for my bad english it´s not my native.

i modded a GTX 770 to a Tesla K10 ( single chip) with the help of this great thread.

i own one of the new GTX 780 6GB Version and i have a nvidia titan black.

Device ID from the GTX 780 6GB is 1007
Device ID from the Titan Black 6GB is 100C

i checked the resistors around the chip.



so for  the gtx 780 the 45K is 15K in my Case, 28,5k is 30k in my case and 25K is 25K
so for  the titan black the 45K 35K in my Case, 28,5k is 30k in my case and 25K is 25K

the screenshot is a 780 3GB ID 1004, can anyone check this resistor on a GTX 780ti Dev ID 100a must be 20K or 25K

so could the missing 4 nipple this resistor ?

1004 = 45K
1007 = 15K
100C = 35K

http://www.guztech.nl/wordpress/index.php/2013/11/researching-nvidia-gpus-geforce-gtx780-and-gtx-titan-to-tesla-k20-and-tesla-k20x/

the guy say, you can reach 101F via softstrap

Thus the first five bits of the device ID can be set in the firmware, meaning that the device ID could be set to values between 0×1000 and 0x101F without modifying any hardware.

so 101F is a Tesla K20. I need only the TCC option, to render via remote, can i eaiely mod the GTX 780 DEV ID 1007 to 1027 ( K40st) or did i get memory problems ?

can anyone confirm this ?

kind regards thunderbolt

[gordan]

Soft strap modding doesn't work on Keplers, at least not for VGA pass through purposes. There is a register that can only be set by the hard strap. If that register doesn't match a white listed ID, drover will refuse to initialise the card.

[thunderbolt]

so i modded today my GTX 780 1007 with the 33K resistor.

now i have a K40st id 1027

update, used to original bios from the GTX 780 and modify the following

DE 10 07 10 Block 794 - 797 to DE 10 27 10

and

DE 10 07 10 Block 7FA20-FA23 to DE 10 27 10

update the checksum and reflash the bios. What i have todo now ?

[gordan]

There shouldn't be anything else to be done. If you have verified that the PCI device ID and the device ID as reported by GPU Caps Viewer both report the same, that's all you can do.

[thunderbolt]

i cant select the k40 in gpu caps viewer, i flashed the original bios back to the k40.

i use this card without any modification on the bios. with TCC it works without problems. is there a problem with the memory size ?

nvidia-smi shows me the correct size, the other tools not

[gordan]

Memory size isn't going to be the problem. I don't know why it doesn't work with 3DS - I'm not a 3DS user so I don't have any ideas.

[thunderbolt]

no you did not understand, all works flawless, i only want to make sure, that the gtx 780 (K40st) can use the full 6GB of ram. nvidia-smi reports the right size, atm no error so far.

[gordan]

The memory size is set in the MCH register on the cad by the binary payload in the BIOS. The BIOS the card ships with will do the right thing to initialize it, and the data in these registers can then be read back by the driver, which will act accordingly. Don't worry about it, if GPU-Z is showing the correct memory size, it's fine.

[oerg866]

Hi there!

First of all, great work!

I have decided to virtualize my windows system, and I wanted to mod my Palit Jetstream GTX 670.

I have removed the R2 resistor. It does not give any picture when booting, but it gets detected in device manager with DEV_118F (Tesla K10) when used as a secondary card. I get an error code 28 when trying to install a driver for that card though.

Do I need to do anything else beside this? I'm kind of reluctant as the PCB is not 100% based on the reference design.

Do I need to edit the firmware for the system to work again? I have an UEFI BIOS. Motherboard is a GIGABYTE GA-X79-UD5.

Cheers!